AI research paths in India
There 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.
In 60 seconds
AI research paths in India
There 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.
The main paths
| Path | What it looks like | Good for | Things to know |
|---|---|---|---|
| Undergraduate research | Working with a professor or a senior student on a project, a thesis, or a summer project | Trying research early and getting a strong letter | Ask early. Show a small piece of work. Expect to start with support tasks. |
| Master's or PhD | Formal research training with an advisor | Deep, long-term research and academic careers | The advisor and the lab's culture matter as much as the institution's name. Talk to current students. |
| Research engineer roles | Building the code, data pipelines and experiments that research needs, in a company or lab | Strong coders who like making research work at scale | Often a route into research without a PhD. Show clean code and reproductions. |
| Open-source contributions | Fixing bugs, adding features, writing evaluations, building datasets in public projects | Learning from real code, getting noticed, building a public record | Start with small, useful issues. Follow the project's contribution guide. |
| Independent research | Doing your own projects and sharing them publicly | Self-learners, working professionals, people outside big cities | Harder without feedback. Join communities, find a mentor, share early. |
Build a public record of your work
- A reproduction write-up with open code (Lesson 2).
- A clean small experiment with honest results (Lessons 3 and 5).
- Contributions to open projects, for example open datasets and models for Indian languages from groups such as AI4Bharat at IIT Madras.
- A preprint on the arXiv or a workshop paper, when you have something complete.
Finding a mentor
- 1
Look close first
Professors, senior students and alumni at your college; people in local meetups and reading groups. - 2
Read their work
Pick one person whose recent papers you have actually read and have a real question about. - 3
Do something small first
Reproduce one of their results, or try an idea from their paper's future work. - 4
Write a short, specific email
Who you are, what you read, what you did, one clear question or request. - 5
Follow up once, politely
If no reply after a couple of weeks, one short follow-up. Then move on without taking it personally.
- "Respected Sir/Madam, I am very passionate about AI. Please give me an internship."
- Same message sent to a hundred professors
- Long CV attached, no link to any work
- "I read your paper on X. I reproduced Table 2 on a smaller dataset and got Y (code link). I noticed Z. Could I work on this under your guidance?"
- Written for one person, about their work
- Short, with a link to your write-up and code
Research integrity: the rules that protect everyone
- Never fabricate or change data or results. Report what you actually got.
- No cherry-picking. Don't hide seeds, datasets or metrics that went badly. Say what you tried.
- No plagiarism. Cite ideas, text, code and figures from others. Copying and rewording without credit is still plagiarism.
- Fair authorship. Credit people who did the work. Discuss authorship early.
- Consent and privacy. Collect data about people only with their informed consent, and protect it. Follow your institution's ethics review process for human data.
- Respect licences. Use data, models and code only as their licences allow.
- Think about harm. Ask how your work could be misused, and write about it where it matters.
- Honest use of AI tools. Follow venue and institution rules, disclose use where asked, and check every claim and citation yourself.
Watch and read more
Lab
A one-page research plan for a small project you can start this month, with a path, a mentor approach and an integrity check.
The problem
Write a one-page research plan. Keep it small enough to finish in about two months of part-time work. Use the headings in the starter.
- Write one research question and a testable hypothesis.
- Name one or two papers it builds on (ones you have read).
- Describe the experiment: data, baselines, ablations, metric, seeds.
- Add a compute plan with total hours and a plan B.
- Say how you will share the result.
- Name one possible mentor type (not a specific person unless you know them) and draft a three-sentence first email.
- Fill in the integrity checklist.
ONE-PAGE RESEARCH PLAN
Question:
Hypothesis (and what result would prove it wrong):
Builds on (1-2 papers I have read):
Data (source, licence, consent if about people):
Baselines:
Ablations:
Metric and seeds:
Compute plan (hours, where, plan B):
Timeline (weeks 1-8):
How I will share it (blog / arXiv / workshop / code):
Path this fits (undergrad / master's / PhD / research engineer / open source / independent):
Mentor approach + draft 3-sentence email:
Integrity checklist:
[ ] all runs will be logged and reported
[ ] settings chosen on validation data only
[ ] data and model licences checked
[ ] consent and privacy handled for any data about people
[ ] others' ideas and code cited
[ ] possible misuse consideredYou are done when
Hard questions
Try to answer before you reveal. If you can answer these, you understood the lesson.
Q1Halfway through, your hypothesis looks wrong. Have you failed?Reveal
No. A clean experiment that disproves your hypothesis is a real result. Check it carefully, then write it up honestly, and look at why it failed. That often points to a better question. What would be a failure is hiding the result or changing the experiment until it "works".
Q2Your plan uses a public dataset scraped from social media. What should you check before using it?Reveal
Check its licence and terms of use, how it was collected, and whether it contains personal data. Think about whether people in it could be harmed or identified by your work. Follow your institution's ethics process if needed, and describe these points in your write-up.
Questions people ask
Do I need a PhD to do AI research?
No. A PhD gives deep training and is often needed for academic jobs and some research roles. But research engineers, open-source contributors and independent researchers also do real research. What counts is careful, honest, shared work.
I am not at an IIT or IISc. Can I still do research?
Yes. Good research happens in many places. Free papers on the arXiv, open models and datasets, free compute tiers and online communities have made it far more open. Build a public record and reach out to people whose work you have read.
How do I find a research problem?
Start from papers you reproduced. Look at their limitations and "future work" sections. Ask where the method fails, for example on Indian languages or low-resource settings. A small, clear question you can answer is better than a huge one you can't.
What if a professor doesn't reply to my email?
This is very common and usually not about you. Professors get many emails. Send one polite follow-up, then approach others. Meanwhile, keep building your public record, which makes the next email stronger.
Lesson test
5 questions. Get 3 right (60%) to pass and complete this lesson.