38 tracks · 335 lessons · free, in your language

The course

Sign in to save your progress and earn certificates. Sign in

Start hereWhat AI actually isBefore we can keep AI safe, we need to know what it is. Short answer: it is a machine that guesses really, really well.Foundations · Lesson 1 · 4 min read

Learning paths

Short routes through the course, one goal each.

Tracks

Open a track to see its units and lessons.

Starts at: Beginner6 lessons2 units

Foundations

Start here if you are new to AI. You will learn what AI and large language models really are, and how a chatbot turns into an agent that can take actions.

What you will be able to do
  • Explain in plain words what AI and a large language model are
  • Tell the difference between a chatbot and an agent
  • Describe how tools, context and memory shape what an agent can do
  • Spot the new risks that appear when several agents work together
Start track →
Show units and lessons

Unit 1

How AI works

What AI is, and what a language model actually does.

  1. Lesson 1: What AI actually isBefore we can keep AI safe, we need to know what it is. Short answer: it is a machine that guesses really, really well.Beginner4 min read🧪 Lab📝 Test
  2. Lesson 2: What a large language model doesAn LLM is autocomplete that swallowed a library. Understanding that one sentence explains most of its strengths and every one of its weaknesses.Beginner4 min read🧪 Lab📝 Test

Unit 2

From chatbot to agent

Tools, memory and teams: how a model starts to act.

  1. Lesson 3: From chatbot to agentA chatbot talks. An agent acts. That one difference is where the entire security field in this guide comes from.Beginner5 min read🧪 Lab📝 Test
  2. Lesson 4: Tools: giving the model handsA tool is any button the model is allowed to press. Click each part of the diagram to see where trust is won and lost.Beginner6 min read🧪 Lab📝 Test
  3. Lesson 5: Context, memory and forgettingThe model has no memory. It has a sheet of paper that gets re-read from scratch every single turn. Once you see that, a whole class of attacks makes sense.Beginner3 min read🧪 Lab📝 Test
  4. Lesson 6: When agents work in teamsOne agent is a risk you can picture. Five agents passing notes to each other is a risk that grows faster than your ability to watch it.Middle3 min read🧪 Lab📝 Test

Starts at: Beginner13 lessons5 units

AI for young minds

For young people aged 10 to 16, and the parents and teachers who help them. You will learn to spot fake pictures, voices and news, stay safe with chatbots and strangers online, use AI to learn (not to copy), see how AI learns from examples, and take charge of your own time and creations.

What you will be able to do
  • After this track you can play "Real or AI?" like a detective and check where a picture, video or message really came from.
  • You can spot scams, fake calls and tricky strangers, and you know exactly who to tell and what to do if something goes wrong.
  • You can use AI as a study buddy that explains and quizzes you, and you know how to check when it is wrong.
  • You can explain in your own words how a computer learns from examples, and why that can sometimes be unfair.
  • You can make things with AI honestly, give credit, and choose how you spend your screen time.
Start track →
Show units and lessons

Unit 1

Real or AI?

Become a fake-spotting detective for pictures, voices and viral posts

  1. Lesson 70: Is it real or made by AI?AI can now make pictures and videos that look real. Learn the clues, and the one habit that works even when the clues disappear.Beginner4 min read🧪 Lab📝 Test
  2. Lesson 71: Voice clones and fake callsAI can copy a voice. Learn how scammers use this, and a simple family plan that beats them.Beginner3 min read🧪 Lab📝 Test
  3. Lesson 72: Check before you forwardFamily groups are full of forwards. Some are true, some are not. Learn a quick check, and how to kindly correct a relative.Beginner3 min read🧪 Lab📝 Test

Unit 2

Safe and smart online

Protect your privacy, spot tricky strangers and know what to do if things go wrong

  1. Lesson 73: Chatbots are not your secret diaryChatbots can be helpful, but they are not a safe place for your private details or your biggest worries.Beginner4 min read🧪 Lab📝 Test
  2. Lesson 74: AI friends, strangers and tricksSome apps offer AI friends, and some strangers online pretend to be someone else. Learn the red flags and what to do.Beginner4 min read🧪 Lab📝 Test
  3. Lesson 75: If someone threatens you onlineIf someone scares you or tries to blackmail you online, it is never your fault. Here is exactly what to do, calmly, step by step.Beginner4 min read🧪 Lab📝 Test

Unit 3

Learn with AI

Use AI to understand more, not to skip the learning

  1. Lesson 76: AI as your study buddy, not your answer keyAI can explain, quiz you and help you plan. But if it does your homework, your brain misses the workout.Beginner4 min read🧪 Lab📝 Test
  2. Lesson 77: When AI is confidently wrongAI can sound very sure and still be wrong. Learn why, and how to catch its mistakes.Beginner3 min read🧪 Lab📝 Test

Unit 4

How AI thinks

Peek inside: how computers learn from examples, and why that can go wrong

  1. Lesson 78: How a computer learns from examplesAI doesn't learn like you do from a teacher's rules. It finds patterns in lots of examples. Try being the AI yourself!Beginner3 min read🧪 Lab📝 Test
  2. Lesson 79: Why AI can be unfairIf AI learns from one-sided examples, it can treat some people unfairly. Learn how this happens and why speaking up matters.Beginner3 min read🧪 Lab📝 Test

Unit 5

Your attention, your creations, your future

Take charge of your screen time, create honestly, and grow skills for tomorrow

  1. Lesson 80: Why you can't stop scrollingApps learn what keeps you watching. Understanding how they work helps you choose how to spend your time.Beginner3 min read🧪 Lab📝 Test
  2. Lesson 81: Make something with AI, and give creditAI can help you make stories, comics, songs and project ideas. Stay the creator, be honest, and never use it to hurt people.Beginner3 min read🧪 Lab📝 Test
  3. Lesson 82: Jobs of the future and skills AI can't replaceAI will change many jobs and create new ones. The skills that matter most are ones you can start building today.Beginner5 min read🧪 Lab📝 Test

Starts at: Beginner8 lessons3 units

AI in everyday life

For anyone with a phone, no background needed. You will notice the AI you already use every day, use chatbots in a smart and honest way, and protect yourself and your family from AI scams, fake videos and privacy traps.

What you will be able to do
  • After this track you can point out the AI features already working in your phone and your apps.
  • You can use a chatbot for study and daily tasks, and check its answers before you trust them.
  • You can spot the common AI scams, and you know what to do in the first minutes if something goes wrong.
  • You can protect your private data and change the key settings in AI apps.
  • You can talk calmly with children and elders about AI and set healthy family habits.
Start track →
Show units and lessons

Unit 1

The AI around you

What is already working in your pocket, and how to use it well

  1. Lesson 83: The AI already in your phoneYou use AI many times a day without noticing. Let us find it, so you can use it better and know its limits.Beginner5 min read🧪 Lab📝 Test
  2. Lesson 84: Why your feed shows what it showsVideos, posts and shopping suggestions are chosen for you by AI. Learn how it picks, why it can trap you, and how to take back control.Beginner4 min read🧪 Lab📝 Test
  3. Lesson 85: Using chatbots well for study and daily tasksA chatbot can be a patient tutor and a handy helper. Used badly, it can make you learn less and believe wrong things. Here is how to use it well and honestly.Beginner5 min read🧪 Lab📝 Test

Unit 2

Staying safe

Scams, fakes and privacy: protect yourself and the people you love

  1. Lesson 86: AI scams: how to spot them and what to doScammers now use AI to copy voices, faces and official-looking messages. The good news: a few simple habits beat almost all of these tricks.Beginner6 min read🧪 Lab📝 Test
  2. Lesson 87: Spotting deepfakes and AI-made imagesAI can make photos, videos and voices of things that never happened. Looking closely helps a little. Checking the source helps a lot more.Beginner5 min read🧪 Lab📝 Test
  3. Lesson 88: Your data and privacy with AI appsWhat you type into an AI app may be stored, read or used to improve it. Learn what never to paste, and which settings to check.Beginner4 min read🧪 Lab📝 Test

Unit 3

AI in your community and family

What AI can and cannot do for health, farming, services and children

  1. Lesson 89: AI in health, farming and government servicesAI is showing up in clinics, fields and public offices. Learn what it can really help with, where it falls short, and why a human expert still matters.Beginner5 min read🧪 Lab📝 Test
  2. Lesson 90: AI, kids and familiesChildren are growing up with AI in their games, videos and homework. Here is how families can talk about it and set healthy habits together.Beginner4 min read🧪 Lab📝 Test

Starts at: Beginner8 lessons3 units

Prompting like a pro

For anyone who uses a chatbot and wants better answers. You will learn to write clear requests, give the right context and examples, improve answers step by step, check them for mistakes, and keep your data safe while you do it.

What you will be able to do
  • After this track you can turn a vague request into a clear prompt with a goal, audience, format and length.
  • You can use examples, roles and constraints to get the style and shape you need.
  • You can work in steps (plan, draft, improve) instead of hoping for a perfect first answer.
  • You can check answers for made-up facts and ask for sources you then verify.
  • You can use AI for everyday work safely, without pasting secrets or following hidden instructions in pasted text.
Start track →
Show units and lessons

Unit 1

Prompt basics

Why wording matters, and the ingredients of a clear request

  1. Lesson 91: What a prompt is and why wording mattersA prompt is simply what you write to an AI. Small changes in wording can change the answer a lot. Here is why.Beginner4 min read🧪 Lab📝 Test
  2. Lesson 92: Be specific: goal, audience, format, lengthFour simple ingredients turn a vague request into a clear one. Learn to add them in seconds.Beginner4 min read🧪 Lab📝 Test
  3. Lesson 93: Give context and examplesThe AI knows nothing about your situation unless you tell it. Background and a few examples can transform the answer.Beginner4 min read🧪 Lab📝 Test

Unit 2

Shaping and checking answers

Control the style, work in steps, and catch mistakes

  1. Lesson 94: Roles, tone and constraintsTell the AI what role to take, how to sound, and what rules to follow. It is like briefing a helper before a job.Middle4 min read🧪 Lab📝 Test
  2. Lesson 95: Step by step: plan, draft, improveBig tasks go better in small steps. Ask for a plan, then a draft, then improvements, checking at each step.Middle4 min read🧪 Lab📝 Test
  3. Lesson 96: Check the answer: sources, verification and hallucinationsAI can invent facts, numbers and even sources that look real. Learn how to ask better, spot warning signs, and verify what matters.Middle5 min read🧪 Lab📝 Test

Unit 3

Prompting for real life

Everyday work, Indian languages, and staying safe

  1. Lesson 97: Prompts for everyday work: email, summaries, tables and translationReady-to-adapt prompts for the tasks people do every day, including working in Indian languages.Middle6 min read🧪 Lab📝 Test
  2. Lesson 98: Prompting safelyKeep secrets out of your prompts, and watch for hidden instructions inside text you paste. Safe prompting protects you, your family and your workplace.Middle5 min read🧪 Lab📝 Test

Starts at: Beginner8 lessons3 units

Python from zero

For anyone aged 12 and up who has never written a line of code. Everything runs right here in your browser, on a phone or a school computer, so there is nothing to install: type, press Run, and see what happens. Step by step you will print messages, store numbers and text, make decisions, repeat work with loops, build your own functions, read error messages calmly and work with small tables of data.

What you will be able to do
  • After this track you can write and run small Python programs in your browser and read what they print.
  • You can store numbers and text in variables, and use if, elif and else to make decisions.
  • You can use for and while loops, lists and dictionaries to work out things like a shop bill.
  • You can write your own functions, and fix common errors by reading the traceback from the bottom up.
  • You can read a small CSV table of marks or train times and calculate totals and averages.
Start track →
Show units and lessons

Unit 1

First steps

Run your first program, store numbers and words, and let the computer make simple decisions

  1. Lesson 99: Your first Python programWrite one line, press Run, and watch the computer do exactly what you said. You will also learn to leave notes for humans with comments.Beginner6 min read🧪 Lab📝 Test
  2. Lesson 100: Variables, numbers and textWhole numbers, decimal numbers and text are different kinds of values. Learn to use each one, convert between them, and print them neatly with f-strings.Beginner7 min read🧪 Lab📝 Test
  3. Lesson 101: Making decisions with ifPrograms become useful when they can choose. Learn to compare values and use if, elif and else to do different things in different cases.Beginner7 min read🧪 Lab📝 Test

Unit 2

Repeating and collecting

Let the computer do repeated work with loops, and keep many values together in lists and dictionaries

  1. Lesson 102: Loops: doing things again and againComputers never get bored. Use for, range and while to repeat steps, and to add up or check many values in one go.Beginner7 min read🧪 Lab📝 Test
  2. Lesson 103: Lists and dictionariesKeep many values together. A list keeps things in order; a dictionary links each name to a value, like a price list.Beginner7 min read🧪 Lab📝 Test

Unit 3

Building blocks

Package your steps into functions, fix errors without fear, and work with small tables of data

  1. Lesson 104: Functions: reusable stepsA function is a named recipe you can use again and again. Learn def, parameters and return, and why returning beats printing.Beginner5 min read🧪 Lab📝 Test
  2. Lesson 105: Reading errors without fearError messages are clues, not punishments. Learn the four errors you will meet most often, and how to read a traceback from the bottom up.Beginner7 min read🧪 Lab📝 Test
  3. Lesson 106: Working with data filesMost real data comes as tables: marks lists, timetables, price lists. Learn to read CSV data with Python's csv module and calculate totals and averages from it.Beginner7 min read🧪 Lab📝 Test

Starts at: Middle12 lessons4 units

Machine learning fundamentals

For learners who know basic Python and want to understand machine learning before large language models. You will train real models with scikit-learn, numpy, pandas and matplotlib, all running in your browser. You will learn to predict numbers, sort things into groups, find groups without labels, clean messy data, and, most importantly, check your results honestly so you do not fool yourself.

What you will be able to do
  • After this track you can explain features, labels, training and prediction in plain words, with a working example.
  • You can train and compare linear regression, logistic regression, decision trees, random forests, nearest neighbours and k-means in scikit-learn.
  • You can split data into training, validation and test sets, spot overfitting, and avoid peeking at the test set.
  • You can read a confusion matrix and choose between accuracy, precision and recall for problems like fraud or rare disease.
  • You can clean a messy table with pandas and run a small end-to-end project with a baseline and an honest evaluation.
Start track →
Show units and lessons

Unit 1

The idea

What it means for a computer to learn from examples, the words everyone uses, and your first model

  1. Lesson 107: What machine learning really doesInstead of writing the rule yourself, you give the computer examples and let it find the rule. Learn what that means, and what it does not mean.Middle6 min read🧪 Lab📝 Test
  2. Lesson 108: Data, features and labelsEvery supervised problem is a table: rows are examples, feature columns are what you know, and the label column is what you want to predict.Middle5 min read🧪 Lab📝 Test
  3. Lesson 109: Your first prediction: linear regressionFit a straight line to data with scikit-learn, use it to predict numbers, and compare it with a simple baseline so you know if it really helps.Middle5 min read🧪 Lab📝 Test

Unit 2

Sorting things into groups

Classification with logistic regression, decision trees, random forests and nearest neighbours

  1. Lesson 110: Classification: sorting into groupsLogistic regression predicts which group a row belongs to, and how sure it is. Learn probabilities, thresholds and why scaling features helps.Middle5 min read🧪 Lab📝 Test
  2. Lesson 111: Decision trees and forestsA decision tree asks a chain of yes/no questions. A random forest asks many slightly different trees and takes a vote.Middle5 min read🧪 Lab📝 Test
  3. Lesson 112: Nearest neighboursTo label a new example, find the most similar examples you already know and let them vote. Simple, strong, and very sensitive to scale.Middle4 min read🧪 Lab📝 Test

Unit 3

Doing it honestly

Spot overfitting, keep a clean test set, and choose measures that do not hide the mistakes that matter

  1. Lesson 113: Overfitting and how to spot itA model that memorises its training examples looks perfect in practice and fails on new data. Learn to see the gap and to control it.Middle5 min read🧪 Lab📝 Test
  2. Lesson 114: Train, validation and test setsUse one part of the data to learn, one to choose, and one, touched only once at the end, to report. Plus cross-validation for small datasets.Middle5 min read🧪 Lab📝 Test
  3. Lesson 115: Measuring a classifier: precision and recallAccuracy can look great while the model misses every case that matters. The confusion matrix, precision and recall show what really happened.Middle5 min read🧪 Lab📝 Test

Unit 4

Real data

Find groups without labels, clean a messy table, and run a full project from made-up data to an honest result

  1. Lesson 116: Clustering: finding groups without labelsk-means finds groups of similar rows when nobody has labelled them. Learn how it works, how to pick the number of groups, and why the groups need a human to interpret.Middle5 min read🧪 Lab📝 Test
  2. Lesson 117: Cleaning messy data with pandasReal tables have missing values, numbers stored as text, spelling variations and duplicate rows. Fix them in a clear order, and keep a record of every change.Middle6 min read🧪 Lab📝 Test
  3. Lesson 118: Project: predict crop yield from rainfallAn end-to-end project on made-up data: explore, split, set a baseline, try a straight line and a curve, and report the result honestly.Middle6 min read🧪 Lab📝 Test

Starts at: Beginner7 lessons4 units

AI for farming

For farmers, farming families, agriculture students and people building agri-startups in India. You will see how AI reads leaf photos, weather records, market prices and satellite images, and where it goes wrong. You will also learn how to build farm tools that respect farmers' knowledge, their language and their data, and that always send serious decisions to a local expert.

What you will be able to do
  • After this track you can explain in plain words what an AI farm app is doing, and when not to trust it.
  • You can explain why a leaf-photo app can be wrong, and why an expert should confirm what it says.
  • You can calculate simple rainfall statistics, a moving average of prices and a crop health index (NDVI) in Python.
  • You can design a farm advice bot that speaks the farmer's language, cites sources and hands hard questions to people.
  • You can ask the right questions about who owns farm data and who gets left out.
Start track →
Show units and lessons

Unit 1

AI in the field

Where AI can help a farmer, and how an app 'looks' at a sick leaf

  1. Lesson 119: How AI can help a farmerAI is a tool that finds patterns in data: photos, weather records, prices. It can save time and give early warnings, but it does not replace the farmer's own knowledge or a local expert.Beginner6 min read🧪 Lab📝 Test
  2. Lesson 120: Spotting crop disease from a photoLeaf-photo apps use image classifiers that learned from many labelled photos. Learn how they decide, why light, background and local varieties can fool them, and why an expert must confirm.Beginner7 min read🧪 Lab📝 Test

Unit 2

Decisions

Using weather, soil and price data wisely, and knowing how uncertain it is

  1. Lesson 121: Weather, soil and sowing decisionsForecasts give chances, not promises. Soil tests tell you what is in your soil today. Learn to read both with their uncertainty, and to summarise past rainfall with a few simple numbers.Middle6 min read🧪 Lab📝 Test
  2. Lesson 122: Market prices and selling smartPrice data can show trends, and a moving average smooths out daily ups and downs. Learn to read prices calmly, and why no tool can tell you tomorrow's price.Middle4 min read🧪 Lab📝 Test

Unit 3

From above

What satellites and drones can see, and the rules for flying

  1. Lesson 123: Drones and satellites over fieldsHealthy plants reflect light in a special way. Satellites and drones use this to make crop health maps, such as NDVI. Learn the idea, compute it yourself, and know the rules before flying a drone.Middle5 min read🧪 Lab📝 Test

Unit 4

Build it right

For builders: advice bots that are safe, and data practices that are fair to farmers

  1. Lesson 124: Building a crop advice bot responsiblyA farm advice bot must speak the farmer's language, show where its answers come from, and hand hard or risky questions to people. It must never guess a pesticide dose.Middle5 min read🧪 Lab📝 Test
  2. Lesson 125: Data, privacy and fairness for farmersFarm apps collect field locations, photos, yields and phone numbers. Learn who should control that data, what real consent looks like, and how tools can leave small farmers and offline villages behind.Beginner5 min read🧪 Lab📝 Test

Starts at: Beginner9 lessons3 units

Stopping AI-powered attacks

For office staff, managers, finance and HR teams, small-business owners and IT generalists. Attackers now use AI to write better phishing, copy voices and faces, and fake invoices and job candidates. You will learn what has really changed, and the simple processes that stop these attacks even when they look perfect.

What you will be able to do
  • After this track you can explain what AI changes for attackers, and what it does not change.
  • You can check a suspicious email by its sender, request and link destination, not by its spelling.
  • You can stop payment fraud and fake-boss requests with callbacks, dual approval and holds on new payees.
  • You can plan a move to phishing-resistant sign-in with passkeys or security keys.
  • You can design awareness training and a reporting culture that fit the AI era.
Start track →
Show units and lessons

Unit 1

The new attacker toolkit

What AI gives attackers, and why the old basics still matter

  1. Lesson 126: How attackers use AI todayAI makes attacks faster, cleaner and more personal. But attackers still need someone to click, pay or share. That is where you can stop them.Beginner5 min read🧪 Lab📝 Test
  2. Lesson 127: AI-written phishing at scalePhishing emails no longer have spelling mistakes. Learn what to check instead: the sender, the request, the pressure and where the link really goes.Beginner4 min read🧪 Lab📝 Test
  3. Lesson 128: Voice clones and deepfake video calls at workA call can sound like your director and a video call can show a familiar face, and both can be fake. Learn the signs, their limits, and the process that works anyway.Beginner5 min read🧪 Lab📝 Test

Unit 2

Fraud that targets organisations

Payment fraud, hiring fraud and the truth about AI-made malware

  1. Lesson 129: Fake bosses, fake vendors: payment fraud with AIBusiness email compromise and fake invoices cost organisations real money. A few simple payment rules stop most of it.Beginner6 min read🧪 Lab📝 Test
  2. Lesson 130: Fake job candidates and hiring fraudAI can polish a fake CV, and someone else can sit the interview. Fake job offers also target people looking for work. Learn the checks that protect both sides.Beginner4 min read🧪 Lab📝 Test
  3. Lesson 131: AI-made malware: hype and realityAI can help attackers write code and lures faster. But most attacks still use known techniques, and the basic defences still work.Middle6 min read🧪 Lab📝 Test

Unit 3

Defences that work

Verification, phishing-resistant sign-in and training for the AI era

  1. Lesson 132: Verify out of band: the callback ruleWhen a request is risky, check it through a different channel you already trust. Here's how to design a verification procedure people will actually use.Beginner5 min read🧪 Lab📝 Test
  2. Lesson 133: Phishing-resistant sign-in: passkeys and security keysOTPs, SMS codes and push approvals can be phished or worn down. Passkeys and security keys work only on the real website. Here's why, and how to roll them out.Middle5 min read🧪 Lab📝 Test
  3. Lesson 134: Training people for the AI era of scamsOld awareness training said 'look for typos'. New training teaches people to check requests, report fast, and never feel ashamed. Here's how to update it.Middle5 min read🧪 Lab📝 Test

Starts at: Beginner19 lessons5 units

Agent security

For anyone who builds or runs AI agents. You will see how agents get attacked, why the old security rules fall short, and which defences actually hold.

What you will be able to do
  • Draw the trust boundary of an agent and find where untrusted text gets in
  • Recognise prompt injection, the lethal trifecta and the quiet ways data leaks
  • Limit damage with sandboxing, least privilege and real human approval
  • Monitor, red-team and respond to incidents in your own agents
  • Know the main documents people cite about AI risk
Start track →
Show units and lessons

Unit 1

Why agents change security

The one idea every later lesson depends on.

  1. Lesson 7: Why the old security rules stop workingThirty years of security wisdom assumes code and data are separate things. Agents put them in the same pot and stir.Beginner5 min read🧪 Lab📝 Test
  2. Lesson 8: The trust boundaryOne idea, drawn once, that the next fifteen modules all depend on. Where does trusted instruction end and untrusted content begin?Beginner6 min read🧪 Lab📝 Test

Unit 2

How attacks work

Injection, leaks and agents with too much power.

  1. Lesson 9: Prompt injection, up closeThe number one risk on every serious list of AI vulnerabilities. Step through a live attack and then switch the defences on.Beginner6 min read🧪 Lab📝 Test
  2. Lesson 10: Indirect injection: the invisible noteThe attacker never speaks to your agent. They just leave a note somewhere your agent will eventually read, and wait.Middle6 min read🧪 Lab📝 Test
  3. Lesson 11: The lethal trifectaThree capabilities that are each perfectly fine on their own. Put all three in one agent and you have built a data-theft machine. Flip the switches and watch.Middle6 min read🧪 Lab📝 Test
  4. Lesson 12: How data actually escapesExfiltration rarely looks like a file upload. It looks like a picture, a link, or a slightly odd search query. Click each channel to see the trick.Middle5 min read🧪 Lab📝 Test
  5. Lesson 13: Excessive agency and blast radiusMost agent disasters are not clever attacks. They are an ordinary mistake meeting a permission nobody thought about. Build an agent below and watch the damage number move.Middle7 min read🧪 Lab📝 Test
  6. Lesson 14: The confused deputyYour agent has powers your user does not. An attacker who cannot open a door themselves simply asks the agent to open it for them.Middle4 min read🧪 Lab📝 Test

Unit 3

Attacks that last and spread

Poisoned tools, poisoned memory and agent-to-agent attacks.

  1. Lesson 15: Supply chain: plugins, MCP servers and skillsInstalling a tool server hands a stranger a text box that writes directly into your agent's brain. Most people install them like browser extensions.Advanced4 min read🧪 Lab📝 Test
  2. Lesson 16: Memory poisoningA normal attack ends when the conversation ends. An attack that reaches long-term memory does not end at all.Advanced5 min read🧪 Lab📝 Test
  3. Lesson 17: Attacks that spread between agentsWhen agents talk to each other, a single poisoned message can travel through a system that has no idea it is under attack.Advanced5 min read🧪 Lab📝 Test

Unit 4

Defences that hold

Sandboxes, approvals, guardrails, identity and logging.

  1. Lesson 18: Sandboxing: the room with no windowsYou cannot stop the agent being fooled. You can decide what room it is standing in when it happens.Middle5 min read🧪 Lab📝 Test
  2. Lesson 19: Keeping a human in the loopA human approval step is your strongest control and the easiest one to ruin. The difference is entirely in how you write the dialog box.Beginner5 min read🧪 Lab📝 Test
  3. Lesson 20: Guardrails, and why they leakFilters, classifiers and safety prompts are worth having. Just be honest about what they are: speed bumps, not walls.Middle6 min read🧪 Lab📝 Test
  4. Lesson 21: Secrets, identity and who the agent really isEvery agent action happens as somebody. Getting that somebody right is the difference between an audit trail and a shrug.Advanced5 min read🧪 Lab📝 Test
  5. Lesson 22: Watching what your agent doesYou cannot investigate what you did not record. Agents produce a lot of activity, and almost none of it is logged by default.Middle5 min read🧪 Lab📝 Test

Unit 5

Test, respond and follow the rules

Attack your own agent, plan for incidents, know the frameworks.

  1. Lesson 23: Red-teaming your own agentAttack it before someone else does. Here is a concrete list you can run this afternoon on a system you own.Advanced7 min read🧪 Lab📝 Test
  2. Lesson 24: When it goes wrong: response and kill switchesAssume an incident. Write the plan now, while nothing is on fire and you can still think clearly.Middle5 min read🧪 Lab📝 Test
  3. Lesson 25: The rules everyone actually citesThree documents come up in every serious conversation about AI risk. Here is what each one is for, in plain language.Middle7 min read🧪 Lab📝 Test

Starts at: Middle11 lessons4 units

Securing AI applications

For developers, AppSec engineers and product teams who build chatbots, RAG search and other AI features. You will map the risks with the OWASP Top 10 for LLM applications, threat model a feature, handle model output and retrieved data safely, control cost and abuse, and test and monitor your app before and after release.

What you will be able to do
  • After this track you can explain each risk in the OWASP Top 10 for LLM applications and say which controls address it.
  • You can threat model an AI feature: draw its data flow, mark trust boundaries, name the assets and write abuse cases.
  • You can treat model output as untrusted input, keep secrets and security rules out of prompts, and enforce permissions on retrieved documents.
  • You can set token budgets, quotas and limits that stop denial-of-wallet attacks and runaway loops.
  • You can build security tests into your pipeline and log and monitor an AI app in production with care for personal data.
Start track →
Show units and lessons

Unit 1

Map the risks

A shared list of AI app risks, and a method to find the ones that matter for your feature

  1. Lesson 135: The OWASP Top 10 for LLM applicationsA shared list of the ten biggest risks in apps built on large language models. Learn each one in a line, and where this course covers it.Middle7 min read🧪 Lab📝 Test
  2. Lesson 136: Threat modelling an AI featureBefore you build, draw how data flows, mark where trust changes, and ask what could go wrong. Here is a practical method for AI features.Middle8 min read🧪 Lab📝 Test

Unit 2

Inputs, outputs and data

Handle model output safely, keep secrets out of prompts, and protect the data behind RAG

  1. Lesson 137: Treat model output as untrusted inputModel output is partly written by whoever wrote the text the model read. Escape it, validate it and allow-list it before any other code uses it.Middle8 min read🧪 Lab📝 Test
  2. Lesson 138: System prompt leakage: prompts are not secretsAssume your system prompt will be read by users. Keep keys and security rules out of it, and enforce what matters in code.Middle6 min read🧪 Lab📝 Test
  3. Lesson 139: Securing RAG: permissions for retrieved documentsRetrieval decides what the model sees. If retrieval ignores who the user is, the model will happily answer from documents they should never read.Middle6 min read🧪 Lab📝 Test
  4. Lesson 140: Vector stores and embedding leaksEmbeddings look like harmless lists of numbers. Treat them, and the vector store that holds them, as carefully as the original text.Advanced6 min read🧪 Lab📝 Test

Unit 3

Cost, abuse and reliability

Stop runaway bills and abuse, and design features that do not make people trust wrong answers

  1. Lesson 141: Unbounded consumption: denial of walletEvery model call costs money. Without limits, one attacker, one bot or one bug in a loop can burn your budget or take the feature down for everyone.Middle6 min read🧪 Lab📝 Test
  2. Lesson 142: Misinformation and overreliance in AI featuresModels can be confidently wrong. The danger grows when your product makes wrong answers look trustworthy, or when people and systems act on them without checking.Middle5 min read🧪 Lab📝 Test

Unit 4

Test and ship

Attack your own app on purpose, keep the fixes in your pipeline, and watch it in production

  1. Lesson 143: AI red teaming: methods and toolsRed teaming is a planned attack on your own AI system to find weaknesses before others do. Learn how to scope it, which tools help, and how to measure and report results.Middle6 min read🧪 Lab📝 Test
  2. Lesson 144: Security testing in the AI pipelineRed teaming finds problems once. Your build pipeline keeps them fixed. Turn findings into tests, scan what you depend on, and re-test whenever the model changes.Advanced6 min read🧪 Lab📝 Test
  3. Lesson 145: Logging and monitoring AI apps in productionYou cannot investigate what you did not record. Log the right things about every AI interaction, protect the personal data inside those logs, and watch for abuse.Middle8 min read🧪 Lab📝 Test

Starts at: Advanced12 lessons4 units

Agentic security in the enterprise

For security architects, platform and identity teams, security leaders and engineers who deploy AI agents at work: assistants with tools, coding agents, browser agents and workflow agents. You already know how agents get attacked. Here you learn how to find every agent in your company, give each one a proper identity, harden the protocols they use, control the riskiest agent types, and run and govern them day to day.

What you will be able to do
  • After this track you can find the AI tools and agents already in use in your company and build an inventory with an owner for each one.
  • You can design identities, scopes and delegated access so no agent ever has more access than the job, or the user, allows.
  • You can harden MCP servers, tool catalogues and agent-to-agent links against poisoning, rug pulls and relayed instructions.
  • You can put controls around browser agents, coding agents and business-workflow agents, and enforce policy outside the model with a gateway.
  • You can respond to an agent incident and run an approval, review and audit process that maps to frameworks such as NIST AI RMF and ISO/IEC 42001.
Start track →
Show units and lessons

Unit 1

Agents everywhere

Find the agents you already have, and give each one a real identity

  1. Lesson 146: Shadow AI and the agent inventoryPeople in your company already use AI tools and agents you never approved. You cannot secure what you cannot see, so the first job is to find them and write them down.Middle7 min read🧪 Lab📝 Test
  2. Lesson 147: Agent identity: non-human identities and least privilegeEvery agent should log in as itself, with short-lived credentials and only the permissions its job needs. Shared human accounts and forever-keys turn every mistake into a mystery.Advanced5 min read🧪 Lab📝 Test
  3. Lesson 148: Delegated access: acting on behalf of usersWhen an agent works for a person, it should carry that person's permissions, narrowed further, never widened. OAuth scopes, consent and audit make that real.Advanced6 min read🧪 Lab📝 Test

Unit 2

Protocols and tools

Secure the pipes agents use: MCP servers, tool catalogues and agent-to-agent links

  1. Lesson 149: Hardening MCP servers and clientsThe Model Context Protocol makes it easy to plug tools into agents. That ease is the risk. Here is how to run MCP in a company without handing out your data.Advanced7 min read🧪 Lab📝 Test
  2. Lesson 150: Tool poisoning and rug pullsA tool's description is read by the model as instructions. Attackers hide orders there, or change a trusted tool after you approved it. Here is how to catch both.Advanced6 min read🧪 Lab📝 Test
  3. Lesson 151: Agent-to-agent protocols and trustAgents from different teams and companies are starting to talk to each other directly. Every message from another agent is untrusted input, however polite it looks.Advanced5 min read🧪 Lab📝 Test

Unit 3

High-risk agent types

Browser agents, coding agents and agents that move money or people's data

  1. Lesson 152: Browser and computer-use agentsAgents that click, type and scroll like a person can reach anything a logged-in browser can. Every page they read is a possible attacker.Advanced6 min read🧪 Lab📝 Test
  2. Lesson 153: Coding agents: secrets, packages and pull requestsCoding agents read your code, run commands and open pull requests. They meet untrusted text in issues and dependencies, and they can reach your secrets. Here is how to use them safely.Advanced6 min read🧪 Lab📝 Test
  3. Lesson 154: Agents in business workflows: payments, email and ticketsAgents in finance, HR and support desks touch money and people's data. Old business controls, like maker-checker and approval limits, work well here if you apply them to agents too.Middle6 min read🧪 Lab📝 Test

Unit 4

Run it safely

Enforce policy at runtime, respond to incidents, and govern agents over time

  1. Lesson 155: Runtime security for agents: policies and gatewaysRules written in a prompt can be talked around. Rules enforced by a gateway between the agent and its tools cannot. Here is how to build that layer.Advanced7 min read🧪 Lab📝 Test
  2. Lesson 156: Incident response for agentic systemsThe earlier lesson covered the first ten minutes. This one turns it into a company playbook: roles, steps, evidence, clean-up and a tabletop exercise to practise it.Advanced6 min read🧪 Lab📝 Test
  3. Lesson 157: Governing agents: approvals, audits and accountabilitySecurity controls need a process around them: who approves an agent, how risky it is, who owns it, and how you check it is still safe months later.Middle6 min read🧪 Lab📝 Test

Starts at: Middle11 lessons4 units

AI for security defenders

For SOC analysts, incident responders, detection engineers, AppSec engineers and students aiming for security jobs. You will learn where AI really speeds up defensive work, where it makes things up or can be tricked, and how to keep humans accountable for every decision.

What you will be able to do
  • After this track you can use AI to summarise alerts, draft queries and write reports, while checking every fact against real evidence.
  • You can draft hunt queries, Sigma rules and YARA rules with AI help, and test them on sample data before trusting them.
  • You can use AI in code review and vulnerability work on systems you own or are authorised to test.
  • You can explain how attackers poison logs and inject instructions into AI tools, and design guardrails against it.
  • You can evaluate AI security tools with a fair pilot, and plan your own path into AI security work.
Start track →
Show units and lessons

Unit 1

The AI-assisted SOC

What AI is good at, triage with evidence, and agents with guardrails

  1. Lesson 158: What AI can and cannot do for defendersAI is a fast helper for reading, writing and explaining. It is not a source of truth. Learn where the line is, and how to handle data safely.Middle6 min read🧪 Lab📝 Test
  2. Lesson 159: Alert triage with AITriage means deciding quickly how serious an alert is. AI can gather context and draft a summary, if you give it the evidence and keep the final call.Middle5 min read🧪 Lab📝 Test
  3. Lesson 160: From copilot to agentic SOCAn assistant answers questions. An agent can run investigation steps by itself. That power needs guardrails, approvals and measurement.Advanced6 min read🧪 Lab📝 Test

Unit 2

Hunt and detect

Hypothesis-driven hunting, detection rules and analysis with AI help

  1. Lesson 161: Threat hunting with AIThreat hunting means searching for attackers who have not triggered any alert. AI can help turn an idea into a query, but you must test it.Middle5 min read🧪 Lab📝 Test
  2. Lesson 162: Detection engineering with AI helpDetection engineers turn attacker behaviour into rules. AI can draft and translate rules fast, but every rule must be tested and reviewed like code.Advanced5 min read🧪 Lab📝 Test
  3. Lesson 163: Phishing and malware analysis with AIAI can summarise suspicious emails, pull out indicators, explain scripts and draft YARA rules. Analyse safely, and never trust its verdict without checking.Middle6 min read🧪 Lab📝 Test

Unit 3

Find weaknesses first

Use AI to find and fix bugs before attackers do, only with permission

  1. Lesson 164: AI-assisted vulnerability discoveryFuzzers and static analysis tools find possible bugs. AI can help point them in the right direction and sort the results, but people confirm and report responsibly.Advanced5 min read🧪 Lab📝 Test
  2. Lesson 165: Secure code review with AIAI can review pull requests for security problems in seconds. Learn what it catches, what it misses, and how to check that a fix really works.Middle6 min read🧪 Lab📝 Test

Unit 4

Trust, adopt and grow

Attacks on AI tools, buying wisely, and building a career

  1. Lesson 166: Attacks on AI security tools: poisoned logs and prompt injectionAn AI analyst reads data that attackers can write. Hidden instructions in logs, emails or file names can trick it. Learn how, and how to defend.Advanced5 min read🧪 Lab📝 Test
  2. Lesson 167: Evaluating and buying AI security toolsMany security products now include AI. Ask the right questions, test on your own data, and compare against a baseline before you buy.Middle6 min read🧪 Lab📝 Test
  3. Lesson 168: Careers in AI securityAI security is growing, and there are many ways in. Learn the main roles, the skills to build, and how to show your work safely and ethically.Middle5 min read🧪 Lab📝 Test

Starts at: Advanced11 lessons4 units

Model and data security

For ML engineers, MLOps and platform teams, security architects and risk and compliance people in companies that train, fine-tune or host AI models. You will learn how models themselves are attacked, how to secure the AI supply chain and training pipeline, how to handle personal data safely, and how to run and prove an AI security programme using well-known frameworks and laws.

What you will be able to do
  • After this track you can explain adversarial examples, data poisoning, backdoors, model extraction and membership inference, and choose realistic defences.
  • You can load models safely, verify their hashes and signatures, and describe a model in an AI bill of materials.
  • You can secure a training and fine-tuning pipeline, from dataset access to release gates and protected weights.
  • You can reduce personal data in prompts, logs and training sets, and explain privacy tools such as differential privacy and federated learning.
  • You can map AI work to NIST AI RMF, ISO/IEC 42001, the EU AI Act and India's DPDP Act, and start a 90-day AI security programme.
Start track →
Show units and lessons

Unit 1

Attacks on models

How the model itself can be fooled, corrupted or copied

  1. Lesson 169: Adversarial examples: fooling a modelA tiny, carefully chosen change to an input can flip a model's answer. Learn why this happens, where it matters, and what defences really achieve.Middle7 min read🧪 Lab📝 Test
  2. Lesson 170: Data poisoning and backdoorsIf attackers can change what a model learns from, they can change how it behaves. Learn how poisoning and hidden triggers work, and how to guard your data.Middle5 min read🧪 Lab📝 Test
  3. Lesson 171: Stealing models and training dataA model behind an API can leak itself and the data it learned from. Learn about model extraction, membership inference and memorisation, and the controls that reduce them.Advanced6 min read🧪 Lab📝 Test

Unit 2

The AI supply chain

Where models and data come from, and how to trust them

  1. Lesson 172: Model files that run code: pickle and safetensorsSome model file formats can run code the moment you load them. Learn why, and how to load models safely.Middle5 min read🧪 Lab📝 Test
  2. Lesson 173: Provenance: signing models and the AI bill of materialsYou cannot trust a model you cannot trace. Learn hashes, signing, build provenance and the AI bill of materials.Advanced7 min read🧪 Lab📝 Test
  3. Lesson 174: Securing the training and fine-tuning pipelineThe pipeline that builds your model is a high-value target. Learn to lock down data, compute, secrets, checkpoints and releases.Advanced5 min read🧪 Lab📝 Test

Unit 3

Privacy for AI

Use personal data only where needed, and protect it where you do

  1. Lesson 175: Privacy-preserving machine learningDifferential privacy, federated learning, anonymisation and synthetic data: what each really protects, and what it costs.Advanced5 min read🧪 Lab📝 Test
  2. Lesson 176: Personal data in prompts, logs and training dataPersonal data leaks into AI systems through prompts, logs and datasets. Learn to minimise it, redact it, limit how long you keep it, and respect people's rights.Middle6 min read🧪 Lab📝 Test

Unit 4

Govern and prove it

Frameworks, laws and a programme that shows your AI is under control

  1. Lesson 177: AI risk frameworks: NIST AI RMF and ISO/IEC 42001Two widely used frameworks help you organise AI risk work and prove it to others. Learn what they are and how security teams use them.Advanced5 min read🧪 Lab📝 Test
  2. Lesson 178: The EU AI Act and India's rules for security teamsWhat the EU AI Act and India's data and cyber rules mean for teams that secure AI, in general terms. Always confirm current rules with legal counsel.Advanced6 min read🧪 Lab📝 Test
  3. Lesson 179: Running an AI security programmePut it all together: inventory, risk tiers, policies, secure development, testing, monitoring, response and reporting, plus a 90-day plan to start.Advanced7 min read🧪 Lab📝 Test

Starts at: Middle8 lessons3 units

AI, work and money

For anyone who works, runs a small business or is choosing a career. It explains how AI changes the tasks inside jobs, how to use it well and honestly, what it really costs, and how to avoid the scams that use the word AI to take your money.

What you will be able to do
  • After this track you can break your own job into tasks and see which ones AI may change
  • You can make a simple, realistic plan to build skills that grow in value
  • You can use AI tools at work or in a small business responsibly, with checks and honesty
  • You can explain why AI is not really free and what it costs to run
  • You can spot the warning signs of AI-themed investment scams and know where to report them
Start track →
Show units and lessons

Unit 1

AI and your job

What changes, what grows in value, and how to use AI well at work

  1. Lesson 180: How AI changes jobsAI rarely replaces a whole job in one go. It changes the tasks inside jobs. Knowing which tasks are exposed helps you plan calmly.Middle5 min read🧪 Lab📝 Test
  2. Lesson 181: Becoming AI-ready at workSome skills grow in value when AI arrives. A small, steady learning plan beats panic.Middle5 min read🧪 Lab📝 Test
  3. Lesson 182: Using AI at work, responsiblyAI can save hours. It can also leak secrets, spread mistakes and damage trust. A few habits keep you on the right side.Middle5 min read🧪 Lab📝 Test

Unit 2

AI in business and the economy

Small business uses, the real cost of AI, and the bigger economic picture

  1. Lesson 183: Growing a small business with AIShops, tuition centres, farmers and freelancers can use AI for marketing, replies and paperwork. It helps most when you know its limits.Middle5 min read🧪 Lab📝 Test
  2. Lesson 184: What AI really costsMany AI apps feel free. Running them uses expensive chips, electricity and people. Someone pays, and it helps to know who.Middle5 min read🧪 Lab📝 Test
  3. Lesson 185: AI and the wider economyAI could make work more productive. Who gains, who loses and how fast is still uncertain. Here is how to think about it clearly.Middle5 min read🧪 Lab📝 Test

Unit 3

Protecting your money and your future

Avoid AI investment scams and explore careers in AI

  1. Lesson 186: AI money scams and investing hypeScammers love the word AI. Learn the warning signs of fake trading bots, pump-and-dump schemes and AI-washing. This is education, not financial advice.Middle5 min read🧪 Lab📝 Test
  2. Lesson 187: Careers in AI, in India and beyondAI careers are not only for coders. Here are the main roles, realistic paths into them and free places to start, with no promises.Middle6 min read🧪 Lab📝 Test

Starts at: Middle8 lessons3 units

AI rules and society

For citizens, students, parents and workers who want to understand the public side of AI. It explains how AI can be unfair, what rights you have over your data, how countries are writing rules, and how AI affects creators, elections, schools and the environment. It gives balanced views on questions that are still being debated.

What you will be able to do
  • After this track you can explain how bias enters AI systems and how teams test for it
  • You can describe your main data rights under India's data protection law and compare them with Europe's
  • You can describe, at a high level, how different regions approach AI rules
  • You can check suspicious content before sharing it, especially around elections
  • You can discuss AI's effects on creators, schools and the environment with balance and care
Start track →
Show units and lessons

Unit 1

Fairness and your rights

How AI can be unfair, and what rights you have over your data

  1. Lesson 188: Bias and fairness in AIAI learns from data made by people, so it can repeat unfair patterns quietly and at scale. Here is how bias creeps in and how teams test for it.Middle5 min read🧪 Lab📝 Test
  2. Lesson 189: Your data, your rightsAI runs on data, often about you. India's Digital Personal Data Protection Act, 2023 gives you rights over your personal data. Here are the basics, compared with Europe's GDPR.Middle7 min read🧪 Lab📝 Test

Unit 2

Laws, ownership and responsibility

How countries regulate AI, who owns creative work, and who answers when things go wrong

  1. Lesson 190: How countries regulate AIDifferent places are choosing different ways to govern AI. The EU uses risk levels; others rely more on existing laws and targeted rules. The rules keep changing.Middle6 min read🧪 Lab📝 Test
  2. Lesson 191: Copyright, creators and AIAI learns from huge amounts of human work. Is that fair to writers, artists and musicians? Many questions are still unsettled. Here is the debate and how creators protect themselves.Middle5 min read🧪 Lab📝 Test
  3. Lesson 192: Who is accountable when AI goes wrongWhen an AI system causes harm, “the computer did it” is not an answer. Accountability means knowing who decides, who checks and how problems are reported and fixed.Middle5 min read🧪 Lab📝 Test

Unit 3

AI in public life

Elections and misinformation, schools and exams, and the environment

  1. Lesson 193: Misinformation, deepfakes and electionsAI makes fake photos, voices and videos cheap and convincing. A few verification habits help protect you, your family and your vote.Middle5 min read🧪 Lab📝 Test
  2. Lesson 194: AI in schools and examsAI can be a patient tutor or a shortcut that stops learning. Here is how students, parents and teachers can use it honestly.Middle5 min read🧪 Lab📝 Test
  3. Lesson 195: AI and the environmentAI runs in data centres that use electricity and often water. It can also help protect the environment. Here are the trade-offs and what is being done.Middle5 min read🧪 Lab📝 Test

Starts at: Beginner10 lessons3 units

AGI

For the curious. What people mean by AGI, how anyone could measure it, what is still missing, and why keeping powerful systems pointed the right way is hard.

What you will be able to do
  • Explain the main meanings of AGI and why people disagree about it
  • Read benchmark claims with healthy doubt
  • Describe the alignment problem and specification gaming with examples
  • Outline the research ideas for keeping advanced systems under control
Start track →
Show units and lessons

Unit 1

What AGI means

Definitions, the capability ladder, and measuring progress.

  1. Lesson 26: What people mean by AGIEveryone argues about whether AGI is close. Almost all of that argument is really an argument about what the letters mean.Beginner5 min read🧪 Lab📝 Test
  2. Lesson 27: The capability ladderIntelligence is not one switch that flips. Click through the rungs to see what changes — and what it means for whoever is supervising.Beginner5 min read🧪 Lab📝 Test
  3. Lesson 28: How would we even measure it?Every benchmark that gets famous gets beaten, and then people say it never really measured intelligence. Here is why that keeps happening.Middle6 min read🧪 Lab📝 Test

Unit 2

Where we are

Why models improved so fast, and the honest gap list.

  1. Lesson 29: Why models suddenly got goodNo single genius idea. Mostly a boring answer — more of everything — plus a few tricks that turned a text predictor into something that follows instructions.Middle5 min read🧪 Lab📝 Test
  2. Lesson 30: What is still missingThe honest gap list. These are the things today's systems cannot do, and each one is a reason to keep humans in the loop.Middle5 min read🧪 Lab📝 Test

Unit 3

The alignment problem

Getting a system to do what you meant, and staying in control.

  1. Lesson 31: The alignment problemGetting a system to do what you meant, not what you literally said. Easy with a calculator. Extremely hard with something that optimises.Middle5 min read🧪 Lab📝 Test
  2. Lesson 32: Specification gaming: finding the loopholeNot malice. Not a bug. The system did exactly what you rewarded, and what you rewarded turned out to have a shortcut.Middle5 min read🧪 Lab📝 Test
  3. Lesson 33: Why almost any goal wants resourcesA strange and important idea: whatever a system is trying to do, a few sub-goals are useful for nearly all of them. That is where the long-term worry comes from.Advanced5 min read🧪 Lab📝 Test
  4. Lesson 34: Looking good in the testThe hardest problem in the field: how do you evaluate a system that can tell it is being evaluated?Advanced4 min read🧪 Lab📝 Test
  5. Lesson 35: Staying in controlIf you cannot fully verify a system, you can still bound it. The research directions that treat control as an engineering problem rather than a philosophical one.Advanced6 min read🧪 Lab📝 Test

Starts at: Beginner18 lessons5 units

Build an LLM

Hands on, with code. You build a small GPT step by step — tokenizer, attention, training loop — then learn how real models are scaled, tuned and served.

What you will be able to do
  • Pick a realistic model size for your budget
  • Write a tokenizer, attention and a complete tiny GPT
  • Train a model that converges and plan compute with scaling laws
  • Fine-tune with instructions, preference training and LoRA
  • Evaluate, shrink and serve a finished model
Start track →
Show units and lessons

Unit 1

Plan and prepare the text

Choose what to build, then turn text into numbers.

  1. Lesson 36: Which LLM can you actually build?There are five honest tiers, separated by three or four orders of magnitude in cost. Pick yours before you write a line of code.Beginner6 min read🧪 Lab📝 Test
  2. Lesson 37: Step 1 · The tokenizerBefore a model sees language, something must chop text into pieces. Get this wrong and everything downstream is quietly worse.Middle6 min read🧪 Lab📝 Test
  3. Lesson 38: Step 2 · Embeddings and positionTurning token ids into vectors, and telling the model what order they came in. Two lookup tables, one subtle idea.Middle5 min read🧪 Lab📝 Test

Unit 2

The transformer

Attention, the block, and a complete tiny GPT.

  1. Lesson 39: Step 3 · Attention, derived slowlyThe one idea the whole field rests on. Three vectors per token, one dot product, one softmax. Click the diagram to walk through it.Middle7 min read🧪 Lab📝 Test
  2. Lesson 40: Step 4 · The transformer blockAttention plus a small feed-forward network plus two normalisations plus two residual connections. Repeat N times. That is the entire model.Middle5 min read🧪 Lab📝 Test
  3. Lesson 41: Step 5 · A complete tiny GPTEvery piece assembled into a model you can actually train tonight. Under 100 lines, and structurally identical to a frontier model.Middle7 min read🧪 Lab📝 Test

Unit 3

Training at scale

Loops that converge, good data, compute budgets and many GPUs.

  1. Lesson 42: Step 6 · The training loop that convergesThe model is the easy part. This module is the one that decides whether your run works or wastes a week.Middle7 min read🧪 Lab📝 Test
  2. Lesson 43: Step 7 · Data is the productArchitecture is nearly free — you can copy it. Data is where models actually differ, and where almost all the work is.Middle6 min read🧪 Lab📝 Test
  3. Lesson 44: Step 8 · Scaling laws and the compute budgetHow to decide model size and dataset size before spending money. Move the sliders and watch the bill.Advanced6 min read🧪 Lab📝 Test
  4. Lesson 45: Step 9 · Training across many GPUsOne model no longer fits on one card. Four ways to split it, and the rule for choosing between them.Advanced6 min read🧪 Lab📝 Test
  5. Lesson 46: Step 10 · Making it fastSame model, same GPUs, three times the throughput. Most training runs leave that on the table.Advanced7 min read🧪 Lab📝 Test

Unit 4

Modern upgrades

What a current transformer does differently.

  1. Lesson 47: Step 11 · Modern architecture upgradesWhat separates a 2019 transformer from a 2026 one. Six changes, each small, together substantial.Advanced6 min read🧪 Lab📝 Test
  2. Lesson 48: Step 12 · Long contextGoing from 4k to 1M tokens is not one trick. It is four, and each one costs something.Advanced6 min read🧪 Lab📝 Test

Unit 5

After pre-training

Instructions, preferences, reasoning, LoRA and shipping.

  1. Lesson 49: Step 13 · Teaching it to follow instructionsA base model completes text. It does not answer questions. Supervised fine-tuning is the step that turns one into the other.Middle6 min read🧪 Lab📝 Test
  2. Lesson 50: Step 14 · Preference optimisationSFT teaches the model what a good answer looks like. Preference training teaches it which of two good answers is better — and that is where character comes from.Advanced7 min read🧪 Lab📝 Test
  3. Lesson 51: Step 15 · Reasoning modelsThe newest chapter. Instead of buying capability with a bigger model, buy it with more thinking at the moment of answering.Advanced6 min read🧪 Lab📝 Test
  4. Lesson 52: Step 16 · LoRA, the path most people takeTrain 0.1% of the parameters, get most of the benefit, on one consumer GPU. This is the module with the highest practical value in the whole track.Middle6 min read🧪 Lab📝 Test
  5. Lesson 53: Step 17 · Evaluate, shrink, serveA model that is not measured is not finished. A model that costs too much to run never ships. Both problems, in one module.Middle6 min read🧪 Lab📝 Test

Starts at: Middle11 lessons4 units

Building AI agents

For developers who know basic Python and have used a chatbot API. You will write an agent loop from scratch, design tools and schemas the model can use well, add planning and memory, write a small MCP server, and learn when several agents or a computer-use agent make sense. Then you will test, debug and run agents with sensible cost, speed and safety limits, and learn when not to build an agent at all.

What you will be able to do
  • After this track you can write a working agent loop with stop conditions, a step limit and clear tool errors, using any model provider.
  • You can design narrow, well-described tools with JSON schemas, validate the model's arguments in code and retry on bad output.
  • You can add planning and memory to an agent, and expose your own tools through a small MCP server with least privilege.
  • You can decide when to use one agent, several agents, a fixed workflow or a single prompt, and explain the trade-offs.
  • You can evaluate an agent with traces and test scenarios, and keep its cost, latency and failures under control.
Start track →
Show units and lessons

Unit 1

Agent basics

The loop, the tools and the structured messages that every agent is made of

  1. Lesson 196: The agent loop in codeAn agent is a loop: the model looks, decides, acts, and looks again. Write that loop yourself in about 30 lines of Python, with stop conditions and a step limit.Middle7 min read🧪 Lab📝 Test
  2. Lesson 197: Designing good toolsThe model can only use your tools as well as you describe them. Learn to name, describe and shape tools so the model picks the right one and recovers from mistakes.Middle7 min read🧪 Lab📝 Test
  3. Lesson 198: Structured outputs and function callingGet machine-readable JSON from the model instead of free text, check it in code against a schema, and retry cleanly when it is wrong.Middle6 min read🧪 Lab📝 Test

Unit 2

Planning and memory

Help agents handle longer tasks: plans they can follow, memory they can use, and tools they can share

  1. Lesson 199: Planning and task decompositionLong tasks go better when the agent breaks them into steps. Compare planning first with deciding step by step, and learn when to change the plan.Middle5 min read🧪 Lab📝 Test
  2. Lesson 200: Agent memory in practiceContext windows fill up. Build the four common kinds of agent memory — a recent window, summaries, a notes file and retrieval — and decide what to forget.Middle6 min read🧪 Lab📝 Test
  3. Lesson 201: Writing your own MCP serverThe Model Context Protocol lets any compatible app use your tools. Learn what a server exposes, how it is structured, and how to keep it to least privilege.Middle7 min read🧪 Lab📝 Test

Unit 3

Many agents and real work

Teams of agents, agents that use a screen, and the wisdom to choose something simpler

  1. Lesson 202: Orchestrating several agentsOne agent can hand parts of a job to others. Learn the orchestrator-worker pattern and handoffs, and when more agents help or just add cost and confusion.Middle6 min read🧪 Lab📝 Test
  2. Lesson 203: Building a computer-use agentSome agents work through a screen: they look at a screenshot, then click and type. Learn the loop, why it must run in a sandbox, and where to ask for confirmation.Advanced7 min read🧪 Lab📝 Test
  3. Lesson 204: When not to build an agentAgents are powerful but slow, costly and less predictable. Often a single prompt or a fixed workflow does the job better. Here is how to choose.Middle5 min read🧪 Lab📝 Test

Unit 4

Ship it

Test, debug and run agents so they are useful, affordable and safe to depend on

  1. Lesson 205: Evaluating and debugging agentsYou cannot improve what you don't measure. Record traces of every run, build test scenarios, measure task success, and replay runs to find what went wrong.Middle7 min read🧪 Lab📝 Test
  2. Lesson 206: Cost, latency and reliability for agentsAgents make many calls, so small costs and delays multiply. Learn budgets, caching, parallel tool calls, timeouts, fallbacks and when to hand over to a person.Middle7 min read🧪 Lab📝 Test

Starts at: Middle10 lessons4 units

RAG and search engineering

For developers who build "chat with your documents", support bots and search. You will learn how retrieval works, from embeddings and chunking to hybrid search, re-ranking and grounded answers with citations. Then you will handle messy real documents, keep the index fresh, use SQL or graphs when they fit better, and measure quality with numbers instead of guesses.

What you will be able to do
  • After this track you can explain when retrieval is a better choice than fine-tuning, and when it is not.
  • You can chunk documents with useful metadata, and combine keyword search and vector search with reciprocal rank fusion.
  • You can add re-ranking and query rewriting, including for Indian-language and transliterated queries.
  • You can prompt for grounded answers with citations, say "not found" honestly, and check claims in code.
  • You can build a golden question set and measure recall@k, MRR and answer faithfulness.
Start track →
Show units and lessons

Unit 1

Retrieval basics

Why we search instead of retrain, how embeddings find similar text, and how to cut documents into good pieces

  1. Lesson 207: Why retrieval beats retrainingA model only knows what it saw in training. Retrieval lets it read your fresh and private documents at answer time, and show where each answer came from.Middle6 min read🧪 Lab📝 Test
  2. Lesson 208: Embeddings and similarity searchAn embedding turns text into a list of numbers, so that texts with similar meaning land close together. Learn how similarity is measured and how search stays fast.Middle6 min read🧪 Lab📝 Test
  3. Lesson 209: Chunking documents wellSearch finds chunks, not whole documents. How you cut the text decides what can be found. Learn size, overlap, splitting by headings and useful metadata.Middle8 min read🧪 Lab📝 Test

Unit 2

Better search

Combine keywords and vectors, re-rank the results, rewrite hard queries, and answer only from what you found

  1. Lesson 210: Hybrid search: keywords plus vectorsKeyword search is great at exact words, codes and names. Vector search is great at meaning. Hybrid search runs both and fuses the results.Middle6 min read🧪 Lab📝 Test
  2. Lesson 211: Re-ranking and query rewritingThe first search casts a wide net. A re-ranker reads each candidate closely and puts the best on top. Query rewriting fixes questions that are vague, short or in a different language or script.Middle8 min read🧪 Lab📝 Test
  3. Lesson 212: Grounded answers with citationsFinding the right text is half the job. The other half is an answer that uses only that text, says "not found" when it must, and shows its sources so people can check.Middle6 min read🧪 Lab📝 Test

Unit 3

Real data

Handle PDFs, scans and tables, keep the index in step with changing documents, and use SQL or graphs when they fit better

  1. Lesson 213: Tables, PDFs and messy documentsReal documents are scanned, multi-column, full of tables and images. Most RAG failures start here, before any search happens.Middle7 min read🧪 Lab📝 Test
  2. Lesson 214: Keeping the index freshDocuments change, get deleted and get new versions. Your index must follow, or the bot will confidently quote rules that no longer exist.Middle7 min read🧪 Lab📝 Test
  3. Lesson 215: Knowledge graphs and structured retrievalSome questions are about counting, filtering and connecting facts, not finding a paragraph. For those, SQL or a graph can beat vector search.Advanced7 min read🧪 Lab📝 Test

Unit 4

Measure it

Turn "it seems better" into numbers you can trust

  1. Lesson 216: Measuring retrieval and answer qualityA RAG system can fail at search or at answering. Measure both: recall@k and MRR for retrieval, faithfulness and correctness for answers, on a golden question set.Middle8 min read🧪 Lab📝 Test

Starts at: Middle9 lessons3 units

Evaluating AI systems

For engineers, product managers and researchers who must decide whether an AI system is good enough to ship, or whether a new model is really better. You will learn what public benchmarks can and cannot tell you, build your own test sets, use human raters and AI judges carefully, and run evaluations in CI, in live A/B tests and for safety.

What you will be able to do
  • After this track you can explain what popular benchmarks measure, why scores can mislead, and how test data leaks into training data.
  • You can define what "good" means for your use case and turn it into measurable metrics before you build.
  • You can build a private test set, write clear labelling guidelines, and measure agreement between raters with Cohen's kappa.
  • You can use an LLM as a judge with a rubric, check it for known biases, and calibrate it against human labels.
  • You can run evals as regression tests, design a fair A/B test, and check safety without making the system refuse too much.
Start track →
Show units and lessons

Unit 1

Why evaluation is hard

Public scores, leaked tests, and deciding what "good" means for you

  1. Lesson 217: Benchmarks and their limitsWhat popular benchmarks such as MMLU, HumanEval, SWE-bench and LMArena actually measure, why they wear out, and why a high score does not mean the model can do your job.Middle6 min read🧪 Lab📝 Test
  2. Lesson 218: Contamination: when the test leakedModels learn from huge amounts of web text, which can include the very test questions used to judge them. Learn the signs, and how held-out and fresh test sets protect you.Middle6 min read🧪 Lab📝 Test
  3. Lesson 219: What to measure for your use caseAccuracy is only one number. Task success, faithfulness, safety, speed, cost and user satisfaction all matter. Decide what "good" means before you build.Middle6 min read🧪 Lab📝 Test

Unit 2

Building evals

Your own test set, AI judges and human raters you can trust

  1. Lesson 220: Building your own test setThe most useful evaluation is a private set of real examples from your task, with clear labels. Here is how to collect, label, size and protect it.Middle7 min read🧪 Lab📝 Test
  2. Lesson 221: LLM as a judgeA language model can grade other models' answers quickly and cheaply, if you give it a clear rubric, check it for known biases and compare it with human judgement.Middle7 min read🧪 Lab📝 Test
  3. Lesson 222: Human evaluation done rightHuman judgement is the gold standard only when it is done carefully: clear rubrics, more than one rater, a check on agreement, and blind comparisons.Middle5 min read🧪 Lab📝 Test

Unit 3

Evals in practice

Run evals on every change, test with real users, and check safety

  1. Lesson 223: Regression tests for prompts and modelsA small prompt edit or a model upgrade can quietly break things that used to work. Run your eval suite automatically on every change, with clear thresholds.Middle6 min read🧪 Lab📝 Test
  2. Lesson 224: A/B tests and online metricsOffline evals tell you what might happen. A/B tests with real users tell you what does happen. Learn guardrail metrics, sample size intuition and novelty effects.Middle7 min read🧪 Lab📝 Test
  3. Lesson 225: Evaluating safety and dangerous capabilitiesSafety evals check two things at once: that the system refuses what it should, and that it does not refuse what it should not. At the frontier, labs also test for dangerous capabilities.Middle6 min read🧪 Lab📝 Test

Starts at: Middle8 lessons3 units

Small and on-device models

For developers and students in India who want useful AI without a big cloud bill or a powerful GPU. You will learn when a small model is the right choice, how to read a model licence, how to run and quantise a model on your own computer, how to distil and fine-tune a small model for one job, and how to put it on a phone or serve it cheaply.

What you will be able to do
  • After this track you can explain when a small model beats a big hosted one, and what you give up in quality.
  • You can read an open-weight model's licence and model card and decide whether it fits your project.
  • You can estimate the memory a model needs, run one locally, and explain what 8-bit and 4-bit quantisation do.
  • You can plan a distillation or fine-tuning project for one narrow task, with an evaluation before and after.
  • You can compare the cost of self-hosting a small model with a hosted API, and pick the cheaper option for your traffic.
Start track →
Show units and lessons

Unit 1

Why small

When a small model is the right tool, and what you are allowed to do with it

  1. Lesson 226: Why small models matterBig models are impressive, but small ones can be cheaper, more private, faster and able to work offline. Learn the real trade-offs so you can choose well.Middle5 min read🧪 Lab📝 Test
  2. Lesson 227: Open-weight models and licencesDownloading a model does not mean you may use it any way you like. Learn the difference between open weights and open source, and how to read a model licence.Middle5 min read🧪 Lab📝 Test

Unit 2

Run and shrink

Run a model yourself, make it smaller, and teach it one job well

  1. Lesson 228: Running a model on your own computerTools like llama.cpp and Ollama let you run open-weight models on a laptop. Learn how much memory you need, when a GPU helps, and what a GGUF file is.Middle7 min read🧪 Lab📝 Test
  2. Lesson 229: Quantisation in practiceQuantisation stores each weight with fewer bits so a model needs less memory. See exactly what 8-bit and 4-bit mean with a tiny numpy example.Middle6 min read🧪 Lab📝 Test
  3. Lesson 230: Distillation: teaching a small modelA big teacher model can help train a small student model for one job. Learn how knowledge distillation and synthetic data work, and the licence rules that can forbid it.Advanced6 min read🧪 Lab📝 Test
  4. Lesson 231: Fine-tuning a small model for one jobFine-tuning teaches a small model one narrow task. Learn when it is worth it, how to prepare data, how LoRA and QLoRA fit on one GPU, and how to measure before and after.Advanced7 min read🧪 Lab📝 Test

Unit 3

Deploy

Put small models on phones and edge devices, and serve them without a big bill

  1. Lesson 232: AI on phones and edge devicesRunning AI on the phone or a small device keeps data local and works offline. Learn the runtimes, the battery and memory limits, and how to update models safely.Middle6 min read🧪 Lab📝 Test
  2. Lesson 233: Serving small models cheaplyServing a model to many users costs money every hour. Learn batching, caching, CPU inference and autoscaling, and how to tell when a hosted API is simply cheaper.Advanced6 min read🧪 Lab📝 Test

Starts at: Middle8 lessons4 units

AI for Indian languages

For Indian developers, linguists, students and startups who want AI to work in Indian languages, not only in English. You will see why these languages are harder for today's AI, measure the cost of Indic text, use open datasets and projects, judge speech recognition, speech synthesis and translation honestly, handle Hinglish and other code-mixed text, and collect language data in a fair and ethical way.

What you will be able to do
  • After this track you can explain why many Indian languages are harder for AI: less digital text, many scripts, dialects, code-mixing and speech-first users.
  • You can measure how characters, UTF-8 bytes and tokens differ for the same sentence in English and an Indian language, and explain what that means for cost and context.
  • You can find open datasets and models for Indian languages, and check their licences before you use them.
  • You can compute word error rate for speech recognition, and judge translations with a rubric, back-translation and metrics while knowing their limits.
  • You can plan language data collection with real consent, fair pay, privacy and fair representation of dialects.
Start track →
Show units and lessons

Unit 1

The challenge

Why Indian languages are hard for AI, what that costs, and the open work you can build on

  1. Lesson 234: Why Indian languages are hard for AIIndia has 22 scheduled languages and hundreds of languages and dialects overall. Here is why AI often works less well in them, and what builders can do.Middle6 min read🧪 Lab📝 Test
  2. Lesson 235: Tokens, scripts and the cost of Indic textTokenizers trained mostly on English often split Indian-language text into more pieces. That means higher cost and less room in the context window. Measure it yourself.Middle6 min read🧪 Lab📝 Test
  3. Lesson 236: Datasets and open projects for Indian languagesYou do not have to start from zero. Learn about open projects such as AI4Bharat and Bhashini, where to find datasets, and why you must always check the licence.Middle6 min read🧪 Lab📝 Test

Unit 2

Speech

Listening and speaking in Indian languages: recognising speech fairly and making voices responsibly

  1. Lesson 237: Speech recognition for Indian languagesTurning speech into text is hard with accents, noise and mixed languages. Learn how word error rate works, write it in Python, and test fairly.Middle7 min read🧪 Lab📝 Test
  2. Lesson 238: Text-to-speech and voice for Indian languagesMaking a computer speak an Indian language naturally is about more than sound quality. Names, numbers, consent for voices and misuse of voice clones all matter.Middle6 min read🧪 Lab📝 Test

Unit 3

Text

Translating and understanding real text: judging translation quality and handling mixed languages

  1. Lesson 239: Machine translation and how to judge itMachine translation is useful and imperfect. Learn to judge it with people, with back-translation and with metrics like BLEU and chrF, and know the limits of each.Middle6 min read🧪 Lab📝 Test
  2. Lesson 240: Code-mixing: Hinglish and friendsPeople often mix languages in one sentence and type them in Roman letters. Learn how to handle mixed input, normalise it carefully, and test with real phrases.Middle6 min read🧪 Lab📝 Test

Unit 4

Do it right

Collecting language data fairly, with consent, and giving back to the communities who share it

  1. Lesson 241: Collecting language data ethicallyLanguage data comes from people. Learn how to collect text and speech with real consent, fair pay, privacy and fair representation, and how to give back.Middle6 min read🧪 Lab📝 Test

Starts at: Middle9 lessons4 units

Multimodal and generative AI

For curious learners and developers who know what a large language model is and want to understand the rest: models that read images, make pictures and video, listen and speak, and create music. You will learn how each kind works in plain words, where it usually fails, how watermarks and content credentials work, and how to send images and audio to an AI API with care for cost and privacy.

What you will be able to do
  • After this track you can explain how images, audio and video are turned into tokens or embeddings that a model can use.
  • You can describe how vision-language models and diffusion models work, and predict the kinds of mistakes they make.
  • You can explain how speech, music and video generation work in general terms, and the consent and copyright questions they raise.
  • You can check an image for content credentials and explain why a missing watermark or credential proves nothing.
  • You can build a small feature that sends images to a multimodal API, with sensible limits on cost and privacy, and use it to draft alt text that a person checks.
Start track →
Show units and lessons

Unit 1

Seeing and understanding

How pictures, sound and video become numbers, and how models learn to read images

  1. Lesson 242: What multimodal meansText is only one way to hold information. Learn how images, audio and video are turned into tokens and embeddings, so one model can work with all of them.Middle6 min read🧪 Lab📝 Test
  2. Lesson 243: How vision-language models seeImage patches, a vision encoder, and CLIP-style matching of pictures and words. Plus the mistakes these models still make with counting, small text and left-right.Middle6 min read🧪 Lab📝 Test

Unit 2

Making images and video

How models turn noise into pictures, and why moving pictures are much harder

  1. Lesson 244: How diffusion models make imagesA diffusion model learns to remove noise, a little at a time. Start from pure static, clean it step by step with your words as a guide, and a picture appears.Middle6 min read🧪 Lab📝 Test
  2. Lesson 245: Video generationA video is many images that must agree with each other over time. Learn why that is much harder and costlier than one image, and where today's models still slip.Middle5 min read🧪 Lab📝 Test

Unit 3

Sound

Machines that listen, speak and make music, and the consent questions that come with voices

  1. Lesson 246: Speech models: listening and talkingSpeech recognition turns voice into text, text-to-speech turns text into voice, and newer assistants talk directly. Learn how each works, why speed matters, and why voices need consent.Middle5 min read🧪 Lab📝 Test
  2. Lesson 247: Music and audio generationModels can now make songs, sound effects and background music from a few words. Learn how they work in general, and the hard questions about artists, consent and copyright.Middle5 min read🧪 Lab📝 Test

Unit 4

Putting it together

Models that mix every modality, how to label what AI makes, and how to build with it

  1. Lesson 248: Any-to-any modelsSome models can take in text, images and sound and also answer with any of them. Learn the general idea, why it is useful, and why it makes testing harder.Middle5 min read🧪 Lab📝 Test
  2. Lesson 249: Watermarks and content credentialsInvisible watermarks such as SynthID and signed content credentials (C2PA) help show where an image, video or sound came from. Learn how they work, and what they cannot prove.Middle5 min read🧪 Lab📝 Test
  3. Lesson 250: Building with multimodal APIsHow to send images and audio to a model API, what it costs, how to protect people's privacy in photos, and how to use it well for accessibility, like writing alt text.Middle7 min read🧪 Lab📝 Test

Starts at: Middle7 lessons4 units

AI for science

For school and college students, science teachers and researchers who want a clear, honest picture of AI in science. You will see what AI really adds, look closely at big results like AlphaFold and machine-learning weather models, meet proof assistants and automated labs, and learn how to use AI in research without fooling yourself or others.

What you will be able to do
  • After this track you can explain three ways AI helps science, and why humans still design, test and judge the work.
  • You can explain what AlphaFold predicts and what it does not, and explore a protein in the public AlphaFold database.
  • You can read a science headline about AI and ask the right questions: what stage is this, what was tested, and what is still unknown.
  • You can explain how proof assistants like Lean check mathematics, and how automated labs choose their next experiment.
  • You can spot common problems like data leakage and fake references, and write an honest statement of how you used AI.
Start track →
Show units and lessons

Unit 1

The idea

What AI actually adds to the way science is done

  1. Lesson 251: What AI adds to scienceAI can find patterns in huge data, run fast approximate simulations and suggest ideas to test. People still design the experiments, test the ideas and judge the results.Middle5 min read🧪 Lab📝 Test

Unit 2

Big results

Protein structures, new molecules and materials, and faster weather forecasts

  1. Lesson 252: Protein structure and AlphaFoldProteins fold into 3D shapes, and shape decides what they do. AlphaFold predicts those shapes from the sequence. Learn what it does, what it does not do, and how to explore its public database.Middle5 min read🧪 Lab📝 Test
  2. Lesson 253: AI for drug and materials discoveryAI can screen huge numbers of candidate molecules and even design new ones. But every candidate still faces a long road of lab tests and, for medicines, clinical trials.Middle6 min read🧪 Lab📝 Test
  3. Lesson 254: Weather and climate modelsMachine-learning weather models learn from decades of past weather and can forecast very fast. Physics-based models still matter, and every forecast carries uncertainty.Middle5 min read🧪 Lab📝 Test

Unit 3

New ways of working

Machine-checked mathematics and labs that plan their own next experiment

  1. Lesson 255: AI and mathematicsProof assistants like Lean check every step of a proof by machine. AI can help find proofs, spot patterns and suggest conjectures, and the checker keeps it honest.Middle6 min read🧪 Lab📝 Test
  2. Lesson 256: AI lab assistants and automated experimentsAI can search the literature, help plan experiments and, in self-driving labs, choose the next experiment for robots to run. Safety and human judgment stay in charge.Middle6 min read🧪 Lab📝 Test

Unit 4

Doing it honestly

How AI can fool researchers, and the habits that keep science trustworthy

  1. Lesson 257: Hype, reproducibility and scientific integrityAI makes it easy to get results that look great and are not real. Learn about leakage, overfitting and fake references, and how honest disclosure and review keep science trustworthy.Middle5 min read🧪 Lab📝 Test

Starts at: Middle6 lessons4 units

Robots and world models

For students and engineers who are curious about AI that acts in the physical world. You will learn why robots lag behind chatbots, the main ways robots learn today (demonstrations, simulation, and models that turn images and words into actions), and what researchers mean by a world model. The track is honest about how much is still unsolved, and ends with what robots could and could not do in India.

What you will be able to do
  • After this track you can explain why tasks that are easy for people, like folding a shirt, are hard for robots, and why robot data is scarce.
  • You can describe imitation learning, behaviour cloning and distribution shift, and say why a robot copying a person can drift off course.
  • You can explain training in simulation, domain randomisation and the reality gap, and run a tiny Q-learning robot in Python.
  • You can explain what vision-language-action models and world models are, in general terms, and where their limits are.
  • You can think clearly about uses, jobs, safety and cost for robots in Indian farms, factories and homes, and write a safety checklist for a home robot.
Start track →
Show units and lessons

Unit 1

The gap

Why a chatbot can write an essay but a robot still struggles to fold a towel

  1. Lesson 258: Why robots lag behind chatbotsChatbots learned from a huge amount of text. Robots have far less data, a messy world, real safety risks and costly hardware. Here is why the gap exists.Middle6 min read🧪 Lab📝 Test

Unit 2

How robots learn

Copying people, practising in simulators, and models that turn pictures and words into actions

  1. Lesson 259: Learning from demonstrationsOne of the most common ways to teach a robot is to show it what to do. Learn how demonstrations are collected, what behaviour cloning is, and why copying goes wrong.Middle6 min read🧪 Lab📝 Test
  2. Lesson 260: Simulation and sim-to-realReal robots are slow and costly to train. Simulators let a robot practise a huge number of times. The hard part is making what it learned work in the real world.Middle7 min read🧪 Lab📝 Test
  3. Lesson 261: Vision-language-action modelsSome newer robot models take camera images and a written instruction, and output the robot's next movements. Learn how they work in general terms, and where they still fall short.Middle6 min read🧪 Lab📝 Test

Unit 3

Understanding the world

Models that predict what happens next, and why many researchers think this matters

  1. Lesson 262: World models: learning how the world movesA world model predicts what will happen next if the robot takes an action. It lets a system imagine before it acts. Learn the idea, the methods and the debate.Middle8 min read🧪 Lab📝 Test

Unit 4

Robots here

Where robots might help in India, and the questions of jobs, safety and cost

  1. Lesson 263: Robots in India: farms, factories and homesWhere could robots be useful in India, and what would it take? A look at possible uses, and the hard questions about jobs, safety and affordability.Middle7 min read🧪 Lab📝 Test

Starts at: Beginner10 lessons3 units

Build a mini robot

For students aged 12 and up, teachers and hobbyists. You start with a small virtual robot that runs in your browser and teach it to drive, follow a line and avoid obstacles. Then you choose low-cost parts, wire a real robot safely and add smooth steering. Finally you give it simple eyes and a voice interface, with safety rules the AI cannot break.

What you will be able to do
  • After this track you can write a robot's decide step in Python and test it in a simulator before touching real parts.
  • You can explain how IR line sensors and an ultrasonic distance sensor work, and why real readings are noisy.
  • You can choose parts for a low-cost two-wheel robot and wire motors, a driver board and sensors safely.
  • You can tune a PID controller so the robot follows a line smoothly.
  • You can add a tiny image classifier and voice commands, with an allow-list so the robot only does safe things.
Start track →
Show units and lessons

Unit 1

A robot in your browser

Meet a small virtual robot, learn what its sensors tell it, and teach it to follow a line and dodge obstacles

  1. Lesson 264: Meet your virtual robotEvery robot runs the same loop: sense, decide, act. Meet a small two-wheel robot that lives in your browser, and drive it to a goal.Beginner7 min read🧪 Lab📝 Test
  2. Lesson 265: Sensors: how a robot notices the worldLine sensors see dark and light, ultrasonic sensors measure distance with sound, and every sensor is a bit noisy. Learn what each one really tells you.Beginner7 min read🧪 Lab📝 Test
  3. Lesson 266: Line follower in codeTwo small light sensors and a few if-statements are enough to follow a curving black line. Write the classic line-follower and learn what to do when the robot loses the line.Beginner6 min read🧪 Lab📝 Test
  4. Lesson 267: Avoiding obstaclesUse the distance sensor to wander around a room full of obstacles without touching any. Learn simple avoidance rules, why robots get stuck, and how to test fairly.Beginner5 min read🧪 Lab📝 Test

Unit 2

Build the real thing

Choose low-cost parts, wire motors and sensors safely, and steer smoothly with PID

  1. Lesson 268: Choosing parts on a budgetA simple two-wheel robot needs a small board, a motor driver, two geared motors, sensors and batteries. Learn what each part does and how to choose without overspending.Beginner7 min read🧪 Lab📝 Test
  2. Lesson 269: Wiring motors and sensors safelyConnect the motor driver, motors and ultrasonic sensor to an Arduino, upload a first test sketch, and test with the wheels off the ground. Safety first, every step.Beginner9 min read🧪 Lab📝 Test
  3. Lesson 270: Smooth steering with PIDA PID controller steers by how far off the line you are, how long you have been off, and how fast the error is changing. Tune P, I and D on the virtual robot until it tracks the line smoothly.Middle6 min read🧪 Lab📝 Test

Unit 3

Add AI

Give your robot a tiny vision model and voice commands, with safety rules the AI cannot break, then build and share

  1. Lesson 271: Giving your robot eyes: a tiny vision modelA camera and a tiny image classifier can tell your robot which way the path turns. Train one on small synthetic images and learn why real-world data matters.Middle6 min read🧪 Lab📝 Test
  2. Lesson 272: Voice commands with a language modelSay "go forward a little" and let a language model turn it into a command. Then make sure the robot only ever accepts a small list of safe commands, whatever the model says.Middle6 min read🧪 Lab📝 Test
  3. Lesson 273: Showcase: plan, build and share your robotPick a small robot project, plan it with a safety and test plan, build it step by step, and share it so others can learn from it.Beginner6 min read🧪 Lab📝 Test

Starts at: Middle7 lessons3 units

TinyML: AI on tiny chips

For makers, engineering students and hobbyists. You will learn what fits on a microcontroller with kilobytes of memory and no operating system, collect and label sensor data, train a tiny classifier, and shrink it with int8 quantisation. Then you will tune an always-on detector for false alarms and misses, and follow the path from a trained model to a working board.

What you will be able to do
  • After this track you can explain what TinyML is, and when a tiny chip is a better home for a model than a phone or the cloud.
  • You can work out a memory budget for a model on a microcontroller: flash for the weights, RAM for the working memory.
  • You can collect and label sensor data, cut it into windows, and train a small classifier on simple features.
  • You can quantise weights to int8, and measure what that saves in size and costs in accuracy.
  • You can choose a detection threshold by measuring false alarms and misses, and plan how to test a model on a real board.
Start track →
Show units and lessons

Unit 1

Tiny but smart

What TinyML is, and how a microcontroller differs from the computers you know

  1. Lesson 274: What TinyML isTinyML means running machine learning models on microcontrollers: tiny, cheap chips that run on very little power. Learn what they can do, and what they cannot.Middle6 min read🧪 Lab📝 Test
  2. Lesson 275: Microcontrollers vs computersA microcontroller has kilobytes of RAM, no operating system and a power budget in milliwatts. Learn how flash, RAM and power decide what model can fit.Middle7 min read🧪 Lab📝 Test

Unit 2

Make a model

Collect and label sensor data, train a tiny classifier, and shrink it with int8 quantisation

  1. Lesson 276: Collecting sensor dataA TinyML model is only as good as its data. Learn how accelerometers and microphones produce data, how to label it, and how to cut it into windows.Middle6 min read🧪 Lab📝 Test
  2. Lesson 277: Training a tiny modelTurn each window into a few useful numbers, such as the mean and the energy, and train a small classifier. Then measure it honestly on held-out data.Middle6 min read🧪 Lab📝 Test
  3. Lesson 278: Shrinking it to fitint8 quantisation stores each weight in one byte instead of four. Do it yourself in numpy, and measure how much size you save and how much accuracy you lose.Middle5 min read🧪 Lab📝 Test

Unit 3

On the device

Always-on detection with the right threshold, and the path from a trained model to a tested board

  1. Lesson 279: Wake words and gesturesAn always-on detector checks again and again, all day. Learn the trade-off between false alarms and misses, and how to choose a threshold.Middle5 min read🧪 Lab📝 Test
  2. Lesson 280: Deploying and testing on a boardFollow the path from a trained model to a program running on a microcontroller. Then test it on real data, and understand the power and privacy benefits.Middle8 min read🧪 Lab📝 Test

Starts at: Middle7 lessons3 units

Drones and self-driving basics

For students and makers who are curious about drones and self-driving machines. You will learn how a drone balances in the air, how sensors like GPS, IMU, cameras and lidar work together, and how a machine works out where it is, plans a path and keeps away from obstacles. Then you will learn how to fly safely and legally in India, and what self-driving cars can and cannot do today.

What you will be able to do
  • After this track you can explain how four rotors lift, balance and steer a drone, and do simple thrust and weight arithmetic.
  • You can describe the strengths and weaknesses of GPS, IMU, cameras and lidar, and combine noisy readings with a simple filter.
  • You can explain why dead reckoning drifts, and fuse it with GPS to get a better position.
  • You can plan a shortest path on a grid map with A*, and add safety margins around obstacles.
  • You can explain India's drone rules in general terms, and give an honest picture of self-driving cars today.
Start track →
Show units and lessons

Unit 1

How they move and sense

How a drone lifts and balances, and how sensors tell a machine what is happening

  1. Lesson 281: How a drone stays in the airFour spinning rotors push air down, and the drone goes up. Learn how thrust, weight and tiny speed changes keep it level, turn it and move it.Middle7 min read🧪 Lab📝 Test
  2. Lesson 282: Sensors: GPS, IMU, cameras and lidarEvery sensor is good at some things and bad at others. Learn what each one measures, where it fails, and why machines combine several of them.Middle8 min read🧪 Lab📝 Test

Unit 2

Finding the way

Working out where you are, planning a route, and keeping a safe distance from obstacles

  1. Lesson 283: Knowing where you areCounting your own steps drifts. GPS jumps around. Combine them and you get something better than either. This is localisation.Middle6 min read🧪 Lab📝 Test
  2. Lesson 284: Planning a pathTurn the world into a grid of free and blocked squares, then search for the shortest route. Learn breadth-first search and A*, the classic path planners.Middle7 min read🧪 Lab📝 Test
  3. Lesson 285: Seeing obstaclesSensors fill a grid map with free, blocked and unknown squares. Then we grow every obstacle by a safety margin, so the robot's body never scrapes past.Middle6 min read🧪 Lab📝 Test

Unit 3

Safe and legal

Flying drones safely and within India's rules, and an honest look at self-driving cars

  1. Lesson 286: Safety, rules and flying legally in IndiaDrones can hurt people and disturb aircraft, so India regulates them. Learn the main ideas of the rules, the airspace zones, and how to plan a safe flight.Middle6 min read🧪 Lab📝 Test
  2. Lesson 287: Self-driving cars: what is real todayDriver assistance is common, and driverless taxis run in a few places under limits. Learn the levels of automation, why rare situations are so hard, and why Indian roads are a special challenge.Middle8 min read🧪 Lab📝 Test

Starts at: Middle7 lessons3 units

Reinforcement learning playground

For students who know basic Python and like games. You will build small games in the browser and write agents that learn to play them from rewards alone: a lock that cracks itself, chai stalls, a treasure hunt, a mango orchard and a go-kart track with a loophole. Then you will see where the same ideas are used in the real world, and why designing the reward is the hardest and most important part.

What you will be able to do
  • After this track you can describe any game as states, actions and rewards, and compute a discounted return.
  • You can write an epsilon-greedy agent for a multi-armed bandit and explain the explore-exploit trade-off.
  • You can implement the Q-learning update, train an agent on a grid game and read the policy it learned.
  • You can design an epsilon decay schedule and compare training runs with numbers, not guesses.
  • You can spot a reward loophole, fix the reward, and explain how RL is used, with care, in robots, recommendations and language models.
Start track →
Show units and lessons

Unit 1

Trial and error

Learning from a score instead of from answers: the loop, the words, and your first bandit

  1. Lesson 288: Learning by trial and errorNobody tells the agent the right answer. It tries something, gets a score, and slowly changes what it does. That is reinforcement learning.Middle6 min read🧪 Lab📝 Test
  2. Lesson 289: States, actions and rewardsEvery RL problem is described with the same few words. Learn them by building a tiny corridor game and scoring it with a discounted return.Middle7 min read🧪 Lab📝 Test
  3. Lesson 290: Multi-armed banditsThe simplest RL game: many choices, each with a hidden chance of reward. Learn epsilon-greedy, the classic way to balance trying and using.Middle6 min read🧪 Lab📝 Test

Unit 2

Learning to plan

Rewards that come later: Q-learning on a treasure hunt, and how much to explore while you learn

  1. Lesson 291: Q-learning step by stepA treasure hunt with rocks and traps. Build the Q-learning update line by line, see why it passes reward backwards, and read the map your agent learns.Middle6 min read🧪 Lab📝 Test
  2. Lesson 292: Exploration versus exploitationExplore too little and you miss the best prize. Explore too much and you play badly for ever. Learn to decay epsilon, and compare runs fairly.Middle6 min read🧪 Lab📝 Test

Unit 3

Beyond games

Rewards that go wrong, and how RL is used, carefully, in robots, apps and language models

  1. Lesson 293: Reward design and its trapsThe agent chases exactly what you reward, not what you meant. Watch a go-kart agent find a loophole, then close it.Middle6 min read🧪 Lab📝 Test
  2. Lesson 294: From games to the real worldHow the same ideas are used for robots, recommendations and training chatbots with human feedback, and why safety matters more outside a game.Middle6 min read🧪 Lab📝 Test

Starts at: Middle8 lessons4 units

Brains and AI

For curious students and learners who want to know how real brains compare with AI. You will see how scientists mapped every neuron in a fruit fly's brain, how a computer model built from that map made predictions that experiments later checked, and how virtual fly bodies walk and fly in simulation. Then you will compare spiking neurons with the neurons in AI models, learn why brains use so little energy, and think about what a wiring map can and cannot tell us.

What you will be able to do
  • After this track you can explain what a connectome is, and what the FlyWire map of a whole fruit fly brain contains.
  • You can build a small made-up connectome in Python and find hub neurons and shortest pathways.
  • You can simulate a leaky integrate-and-fire neuron and explain how it differs from an artificial neuron.
  • You can explain, with a toy calculation, why sparse, event-driven activity saves energy, and what neuromorphic chips try to copy.
  • You can say clearly what a wiring diagram alone does not tell you, and name the ethical questions around brain data.
Start track →
Show units and lessons

Unit 1

Mapping brains

What a wiring map of a brain is, and how a whole fruit fly brain was mapped

  1. Lesson 295: What a connectome isA connectome is a map of every neuron in a nervous system and the connections between them. Learn what such a map shows, and how scientists study it as a network.Middle6 min read🧪 Lab📝 Test
  2. Lesson 296: FlyWire: mapping a whole fly brainIn 2024 the FlyWire team published a complete wiring map of an adult fruit fly brain. Learn how it was built with electron microscopes, AI and many human proofreaders.Middle5 min read🧪 Lab📝 Test

Unit 2

Brains in a computer

Turning a wiring map into a working model, and giving a model fly a virtual body

  1. Lesson 297: Simulating a brain from its wiringResearchers built a computer model of the whole fly brain from the FlyWire map. It predicted which neurons drive behaviours such as feeding, and experiments confirmed some predictions.Middle5 min read🧪 Lab📝 Test
  2. Lesson 298: A virtual fly bodyA brain needs a body. Researchers have built detailed physics simulations of a fly's body that can walk and fly, controlled by trained neural networks.Middle5 min read🧪 Lab📝 Test

Unit 3

Brains versus AI

How real neurons and AI neurons differ, and why brains do so much with so little power

  1. Lesson 299: Spiking neurons vs artificial neuronsReal neurons talk in brief electrical spikes over time. Neurons in most AI models pass smooth numbers. Build a simple spiking neuron and see the difference.Middle5 min read🧪 Lab📝 Test
  2. Lesson 300: Why brains use so little energyThe human brain runs on about 20 watts. Learn the main ideas that make brains so efficient: sparse spikes, events instead of clocks, and memory right next to computing.Middle5 min read🧪 Lab📝 Test

Unit 4

What next

Chips inspired by brains, and what AI can and cannot learn from wiring maps

  1. Lesson 301: Neuromorphic chipsNeuromorphic chips are computer chips designed to work more like brains: spiking neurons, events instead of a clock, and memory close to computing. Learn what they are good for, and their limits.Middle4 min read🧪 Lab📝 Test
  2. Lesson 302: What AI can learn from brainsBrains have inspired AI before, and may again. Learn which ideas are being borrowed, what a wiring diagram alone does not tell you, and the ethical questions around brain data.Middle6 min read🧪 Lab📝 Test

Starts at: Middle6 lessons3 units

Quantum computing and AI

For curious students who know a little algebra. You will learn what qubits, superposition, measurement and entanglement really mean, without myths. You will build a tiny quantum computer simulator with numpy, and then look honestly at what quantum computers might, and might not, do for AI and for security.

What you will be able to do
  • After this track you can explain the difference between a bit and a qubit, and why measuring a qubit gives only 0 or 1.
  • You can turn amplitudes into measurement probabilities and simulate measurements in Python.
  • You can explain entanglement with a Bell state, and why it cannot send messages faster than light.
  • You can simulate small circuits with Hadamard, X and CNOT gates as matrices, and explain why big simulations need 2^n numbers.
  • You can separate proven quantum speed-ups from open questions, and explain why post-quantum cryptography matters today.
Start track →
Show units and lessons

Unit 1

The basics

Qubits, superposition, measurement and entanglement, explained with numbers instead of mystery

  1. Lesson 303: Bits and qubitsA bit is 0 or 1. A qubit is described by two numbers called amplitudes, but when you look at it you still get only 0 or 1. Learn what that really means.Middle7 min read🧪 Lab📝 Test
  2. Lesson 304: Superposition and measurementSuperposition means a state with amplitudes for several outcomes. Measurement turns amplitudes into probabilities by squaring their size. Simulate it in numpy.Middle6 min read🧪 Lab📝 Test
  3. Lesson 305: Entanglement without the mysteryTwo entangled qubits give perfectly matched results, even far apart. Learn what the Bell state is, what the correlations mean, and why no message travels faster than light.Middle5 min read🧪 Lab📝 Test

Unit 2

Try it

Build a tiny quantum computer simulator in numpy and see why big ones are hard to simulate

  1. Lesson 306: Simulating a tiny quantum computerA quantum state is a vector of amplitudes, and gates are matrices. Build Hadamard, X and CNOT in numpy, make a Bell state, and see why n qubits need 2ⁿ numbers.Middle6 min read🧪 Lab📝 Test

Unit 3

Quantum and AI

What quantum computers might do for machine learning, what is proven, and why security teams act now

  1. Lesson 307: What quantum computers might do for AIQuantum machine learning is an active research area with interesting ideas and serious open problems. Learn the main ideas, the data loading problem, and what is not yet known.Middle6 min read🧪 Lab📝 Test
  2. Lesson 308: Hype check: where quantum standsToday's quantum computers are small and noisy. Learn what is proven, why error correction is the big goal, and why post-quantum cryptography matters for security right now.Middle7 min read🧪 Lab📝 Test

Starts at: Advanced5 lessons2 units

Toward AGI

For builders who finished the LLM track. The pieces a bigger model will not give you — memory that grows, grounding and self-improvement — and where the brakes belong.

What you will be able to do
  • Explain what scaling alone will not give you
  • Design memory that builds up over time
  • Connect a system's predictions to real outcomes
  • Place safety brakes in a self-improving system
Start track →
Show units and lessons

Unit 1

Beyond a bigger model

Memory and grounding: the missing pieces.

  1. Lesson 54: What a bigger LLM will not give youYou have now built one. So here is the engineering question honestly: which missing pieces are scale problems, and which are architecture problems?Advanced5 min read🧪 Lab📝 Test
  2. Lesson 55: Memory that actually accumulatesFour kinds of memory, only one of which most systems implement. Building the other three is the most tractable step toward systems that improve with use.Advanced7 min read🧪 Lab📝 Test
  3. Lesson 56: Grounding: finding out you were wrongA model trained on text learns what people say happens. Grounding is connecting a prediction to a consequence — and it is the loop most AI systems still do not close.Advanced5 min read🧪 Lab📝 Test

Unit 2

Closing the loop safely

Self-improvement, and the full stack with its brakes.

  1. Lesson 57: Systems that improve themselvesClose the loop and the system gets better on its own. This is the most exciting architecture in the guide and the one that most needs a hand on the brake.Advanced6 min read🧪 Lab📝 Test
  2. Lesson 58: The reference stack, with the brakes markedEverything from both tracks in one architecture. Click any layer to see what it does and where the control point sits.Advanced7 min read🧪 Lab📝 Test

Starts at: Beginner8 lessons3 units

Build an AGI

Hands on, with code. Build a small autonomous loop that plans, writes code, runs it and fixes its own mistakes — then make it safe enough to trust.

What you will be able to do
  • Explain the three tiers of digital cognition
  • Build a plan, code, run and fix loop around the OODA cycle
  • Make the verifier and the sandbox trustworthy
  • Keep working memory useful over hundreds of steps
  • Mentor others to build such systems responsibly
Start track →
Show units and lessons

Unit 1

The architecture

Why a checker turns a text generator into a reliable system.

  1. Lesson 61: Three tiers of digital cognitionA model that talks, a model that checks itself, and a system that acts on its own. Most confusion about AGI comes from mixing these three up.Beginner7 min read🧪 Lab📝 Test
  2. Lesson 62: The model proposes, the compiler disposesThe single sentence that turns an unreliable text generator into a system you can trust with real work.Middle6 min read🧪 Lab📝 Test
  3. Lesson 63: The OODA loop and the four pillarsThe architecture. Four software components in a continuous cycle — and the reason it is a cycle rather than a pipeline.Middle7 min read🧪 Lab📝 Test

Unit 2

Build it and harden it

The mini-AGI engine, a better verifier and a real sandbox.

  1. Lesson 64: Hands on: the mini-AGI engineOne file, about a hundred lines. It plans, writes code, runs it, reads the error, and fixes itself. Everything above, made real.Middle8 min read🧪 Lab📝 Test
  2. Lesson 65: The bug hiding in your verifierThe loop above declares victory as soon as the code runs without crashing. Running is not the same as being right, and that gap is where self-improving systems go wrong.Advanced7 min read🧪 Lab📝 Test
  3. Lesson 66: Making the executor a real sandboxThe tutorial's executor runs attacker-influenceable code on your machine with your credentials. Here is what has to change before it touches anything you care about.Advanced7 min read🧪 Lab📝 Test

Unit 3

Scale it and pass it on

Memory for long runs, and teaching the next builders.

  1. Lesson 67: Working memory that survives the loopA growing string is fine for three iterations and useless for three hundred. Splitting state from retrieval is what lets an agent work for hours instead of minutes.Advanced7 min read🧪 Lab📝 Test
  2. Lesson 68: Teaching this to the next buildersHow to mentor someone — a curious teenager or a junior engineer — so they build systems that matter instead of another wrapper around a chat box.Beginner6 min read🧪 Lab📝 Test

Starts at: Advanced8 lessons4 units

Inside the model: interpretability

For learners who built a tiny GPT and now want to know what is going on inside it. You will learn the main tools of interpretability research, from linear probes to sparse autoencoders, circuits and activation steering, and try each one in small numpy experiments. You will also learn what these tools cannot yet do, and how they fit into AI safety.

What you will be able to do
  • After this track you can explain the difference between testing a model's behaviour and understanding its mechanism.
  • You can train a linear probe on activations and explain why a working probe does not prove the model uses that information.
  • You can explain polysemantic neurons and superposition, and train a tiny sparse autoencoder that recovers hidden features.
  • You can describe activation patching, induction heads and activation steering, and run toy versions of each.
  • You can judge claims about interpretability for safety: what is a real result, what is a hope, and what is still unsolved.
Start track →
Show units and lessons

Unit 1

Why and how

Why we look inside models, and the first two tools people reach for

  1. Lesson 309: Why look inside a modelTesting what a model does tells you a lot, but not why it does it. Interpretability tries to read the mechanism itself.Advanced6 min read🧪 Lab📝 Test
  2. Lesson 310: Probing: what does a layer know?Train a simple classifier on a layer's activations to see what information is there. Then learn why 'it is there' is not the same as 'the model uses it'.Advanced5 min read🧪 Lab📝 Test
  3. Lesson 311: Attention maps and their limitsAttention weights show where a head looks. They do not show what it takes from there, or what the model does with it.Advanced6 min read🧪 Lab📝 Test

Unit 2

Features

What models seem to represent, why single neurons are hard to read, and how to untangle them

  1. Lesson 312: Features and superpositionOne neuron often responds to many unrelated things. A leading explanation is that models pack more features than they have dimensions.Advanced6 min read🧪 Lab📝 Test
  2. Lesson 313: Sparse autoencodersLearn a large dictionary of sparse features from a model's activations, so that each activation is explained by a few readable pieces.Advanced6 min read🧪 Lab📝 Test

Unit 3

Mechanisms

From what a model represents to how it computes: causal tests, circuits and steering

  1. Lesson 314: Circuits: tracing a behaviourFind which parts of a model cause a behaviour by swapping activations between runs. Induction heads are the classic worked example.Advanced6 min read🧪 Lab📝 Test
  2. Lesson 315: Steering models with activationsAdd a chosen direction to a model's activations while it runs, and its behaviour shifts. Useful for research, and a reminder of how fragile safety training can be.Advanced6 min read🧪 Lab📝 Test

Unit 4

So what

What interpretability can and cannot do for AI safety today

  1. Lesson 316: Interpretability for safety: promise and limitsMany hope that looking inside models will one day catch deception or hidden goals. Here is what is real today, what is a hope, and how it fits with evaluations.Advanced6 min read🧪 Lab📝 Test

Starts at: Advanced7 lessons3 units

Frontier alignment research

For advanced learners who already know why alignment is hard and want to know how researchers actually work on it. You will learn the main research methods, what each one is trying to show, where people disagree, and how to choose and plan a small alignment project of your own.

What you will be able to do
  • After this track you can turn a vague worry about AI behaviour into a testable research question with a clear prediction.
  • You can explain scalable oversight, Constitutional AI, debate and weak-to-strong generalisation, including what each can and cannot show.
  • You can describe how dangerous capability evaluations and responsible-scaling-style policies work, and the main criticisms of them.
  • You can explain what a model organism of misalignment is and design one safely on paper.
  • You can choose a tractable alignment research problem and write a one-page proposal for it.
Start track →
Show units and lessons

Unit 1

Methods

How researchers turn alignment worries into experiments, and four families of methods for supervising strong models

  1. Lesson 317: From concepts to research methodsYou know the worries: loopholes, power-seeking, looking good in the test. Research starts when a worry becomes a question that an experiment or an argument can answer.Advanced7 min read🧪 Lab📝 Test
  2. Lesson 318: Scalable oversightHow do you check the work of a system that may know more than you? Scalable oversight is the family of ideas for supervising AI when human judgement alone is not enough.Advanced7 min read🧪 Lab📝 Test
  3. Lesson 319: Constitutional and rule-based trainingInstead of rating millions of answers by hand, write down the principles and let AI feedback apply them. Learn how Constitutional AI works, why it helps, and where it falls short.Advanced7 min read🧪 Lab📝 Test
  4. Lesson 320: Debate and weak-to-strong generalisationTwo research ideas for supervising stronger systems: let AIs argue while a human judges, and study whether a weak supervisor can train a stronger model well. Learn what these experiments can and cannot show.Advanced6 min read🧪 Lab📝 Test

Unit 2

Testing for trouble

Measuring risky capabilities before release, and building controlled examples of failure to study them safely

  1. Lesson 321: Dangerous capability evaluationsBefore a powerful model is released, teams test whether it can do things that could cause serious harm. Learn how these evaluations are designed, why getting the most out of the model matters, and how thresholds link results to decisions.Advanced7 min read🧪 Lab📝 Test
  2. Lesson 322: Model organisms of misalignmentBiologists study diseases in fruit flies and mice. Alignment researchers deliberately build small, controlled examples of misaligned behaviour so they can study it safely before it appears in the wild.Advanced6 min read🧪 Lab📝 Test

Unit 3

Your turn

Choosing a research problem you can actually make progress on, and planning it

  1. Lesson 323: Choosing an alignment research problemGood research starts with a good problem. Learn how to judge taste, tractability and feedback loops, find and read papers, work with little compute, collaborate, and think about the safety of publishing.Advanced7 min read🧪 Lab📝 Test

Starts at: Middle6 lessons3 units

How to do AI research

For college students, early researchers and self-learners in India who want to do real AI research, not only read about it. You will learn to read a paper quickly, reproduce a result, design a fair experiment, work with little compute, write up what you found, and choose a path into research. No famous lab or big budget is needed to start.

What you will be able to do
  • After this track you can read an AI paper in three passes and judge whether its evaluation is strong or weak.
  • You can reproduce a published result on a small scale, log every difference, and explain a mismatch honestly.
  • You can design an experiment with a clear hypothesis, fair baselines, ablations, several random seeds and a clean train, validation and test split.
  • You can plan experiments that fit a small compute budget and write up your results with clear figures, honest limitations and open code.
  • You can choose a realistic research path in India, approach a mentor well, and follow the basic rules of research integrity.
Start track →
Show units and lessons

Unit 1

Read and reproduce

Read papers fast and well, then check that a result really holds by running it yourself

  1. Lesson 324: Reading a paper in 30 minutesYou do not need to read every word of a paper to understand it. Read it in three passes, take notes in a fixed template, and learn to spot weak evaluation.Middle8 min read🧪 Lab📝 Test
  2. Lesson 325: Reproducing a resultThe best way to truly understand a paper is to make its result appear on your own computer. Start small, match one number, and log every difference.Middle8 min read🧪 Lab📝 Test

Unit 2

Run experiments

Design experiments that give trustworthy answers, and run them with the compute you actually have

  1. Lesson 326: Designing a clean experimentA good experiment answers one clear question fairly. Learn hypotheses, baselines, ablations, random seeds and clean data splits, with a small Python example.Middle9 min read🧪 Lab📝 Test
  2. Lesson 327: Compute on a budgetYou can do real research without a big GPU cluster. Start small, use free and academic resources wisely, write efficient code, and know when you truly need more.Middle7 min read🧪 Lab📝 Test

Unit 3

Share and grow

Write up your work honestly, share it openly, and find a path and mentors in India

  1. Lesson 328: Writing and sharing your resultsResearch that nobody can read or check does not help anyone. Learn the structure of a paper, clear figures, honest limitations, and how to share through preprints, code, blogs and workshops.Middle9 min read🧪 Lab📝 Test
  2. Lesson 329: AI research paths in IndiaThere is more than one road into AI research. Compare the main paths, learn how to find and approach mentors, and understand the rules of research integrity.Middle7 min read🧪 Lab📝 Test

Starts at: Middle6 lessons3 units

Capstone projects

For learners who have finished some lessons and now want to build something real, alone or with friends or a class. You will choose a small problem that matters to people near you, plan it on one page, and build a grounded chatbot or a scam detector. Then you will test it with real people safely and with their consent, and share it honestly with its limits.

What you will be able to do
  • After this track you can choose a project that solves a real problem for real people, and cut it to a size you can finish in two weeks.
  • You can write a one-page plan with a problem, users, data, a success measure, risks and a timeline.
  • You can build a chatbot that answers only from your own trusted content, cites it, and says "not found" when unsure.
  • You can build a rule-based scam detector and measure its misses and false alarms with labelled examples.
  • You can run a small pilot with consent, protect testers' privacy, and publish a demo and README with honest limits.
Start track →
Show units and lessons

Unit 1

Plan

Pick a real problem, cut it down to size, and write the whole plan on one page

  1. Lesson 330: Choosing a project that mattersThe best first project solves a small, real problem for people you can actually talk to. Learn how to find one and cut it down to two weeks of work.Middle6 min read🧪 Lab📝 Test
  2. Lesson 331: Plan it in one pageA one-page plan names the problem, the users, the data, how you will measure success, the risks and the timeline. It keeps a group focused and makes the project easy to explain.Middle6 min read🧪 Lab📝 Test

Unit 2

Build

Two starter builds: a grounded chatbot for your language, and a scam detector for your family

  1. Lesson 332: Build a chatbot for your languageA useful small chatbot answers only from content you trust, cites it, handles how your users really type, and says "not found" instead of guessing.Middle6 min read🧪 Lab📝 Test
  2. Lesson 333: Build a scam detector for your familySimple rules and a score can flag many common scam messages. Learn to measure what it catches, what it misses and how often it raises a false alarm.Middle7 min read🧪 Lab📝 Test

Unit 3

Ship

Test with real people safely and with consent, then share your work honestly

  1. Lesson 334: Test it with real peopleA small pilot with a few real users teaches more than weeks of building alone. Do it with consent, collect only what you need, measure, and improve.Middle6 min read🧪 Lab📝 Test
  2. Lesson 335: Share it and get feedbackA short demo, a clear README with honest limits, and a suitable licence turn your project into something others can learn from, and something you can show.Middle7 min read🧪 Lab📝 Test

Starts at: Beginner3 lessons2 units

Put it to work

Bring it all together. Talk honestly about AI timelines, check an agent before it goes live, and review every key term before the final exam.

What you will be able to do
  • Talk about AI timelines honestly, without hype or dismissal
  • Run a pre-flight checklist before an agent touches real systems
  • Review the key terms and test yourself with the final exam
Start track →
Show units and lessons

Unit 1

Use what you learned

Honest conversations and a checklist you can use today.

  1. Lesson 59: How to argue about timelinesYou will be asked "so when does AI take over?" at dinner. Here is how to answer honestly without sounding like either a hype merchant or a cynic.Beginner5 min read🧪 Lab📝 Test
  2. Lesson 60: The pre-flight checklistEverything in this guide, compressed into things you can tick off before an agent touches anything real. Print it. Use it.Middle4 min read🧪 Lab📝 Test

Unit 2

Final review

Every term in one place, then the final exam.

  1. Lesson 69: Glossary and final examEvery term in one place, then fifteen questions drawn from across the guide. Get sixteen right and you have genuinely understood this.Beginner5 min read🧪 Lab📝 Test

Review and test yourself