AI and robotics internships in India — how to actually get one

We read a lot of intern applications. The rejection is usually decided in about thirty seconds, and almost never for the reason the applicant assumes.

It is rarely about the college. It is rarely about CGPA. It is almost always that nothing in the application shows evidence of the applicant having done anything difficult.

What the pile looks like

A typical application: a one-page CV, a list of six certificates, three coursework projects with names like “Movie Recommendation System”, and a cover letter explaining that the applicant is passionate about AI.

Now consider it from the other side of the table. The team is trying to answer one question: if we give this person a hard, open-ended problem and three months, will something useful come back?

Certificates do not answer that question. Coursework does not answer it, because everyone in the batch did the same project. “Passionate about AI” actively hurts, because every application says it.

What actually answers the question

Something you built that nobody assigned you. Scale is irrelevant. A small tool you made because something annoyed you says more than a large project you were graded on. It demonstrates that you start things without being told to — which is most of what a research internship requires.

A paper you implemented from the paper. Not from a tutorial, not from a repo you cloned. Pick a method, implement it from the description, write up what you got and where you disagreed with the reported numbers. This is close to the highest-signal thing an undergraduate can put in an application, and very few do it.

A write-up of something that failed. A post explaining an approach you tried, why you expected it to work, and why it did not, is a strong signal. It shows you can reason about your own work rather than only celebrate it. Most applicants hide their failures; the ones who explain them stand out immediately.

Depth in one thing. A candidate who knows one area properly beats a candidate with shallow exposure to six. If you have spent a year on computer vision and can talk about the failure modes of the methods you used, that is a conversation. A list of twelve frameworks is not.

The application itself

Lead with the work, not the adjectives. The first line should say what you have built or studied, with a link. Not where you study, and not what you are passionate about.

Write four sentences about this specific team. Reference something they actually published or shipped, and say what you found interesting or questionable about it. This takes fifteen minutes and puts you ahead of ninety percent of applicants, because ninety percent send the same message to forty organisations.

Link things that open. A GitHub with a README a stranger can follow. A deployed demo. A PDF that loads. Anything that requires a request for access will not be chased.

Be specific about availability. “Available from January, 30 hours a week, for six months” is useful. “Flexible” means the team has to work it out, and they will not.

What you do not need

• A top-tier college. It helps at some places; at many deeptech teams it is close to irrelevant next to a good portfolio.
• Prior publications. Most research interns arrive with none.
• A perfect CGPA. It is a weak filter at best, and a low one is easily outweighed by evidence of real work.
• Every technology in the listing. Job descriptions are wishlists. If you meet most of it and can demonstrate you learn quickly, apply.

Where to look

• Research labs at IITs, IISc and IIITs — many take external interns, and a direct, specific email to a professor whose papers you have actually read works better than any portal.
• Deeptech startups — smaller teams, more responsibility, and usually a faster path from idea to something shipped.
• Open source — contributing to a project used by the field is itself a credential, and the maintainers are often the people hiring.
• Programmes like FOSSEE, Google Summer of Code and similar — structured, competitive, and a genuine signal.

On unpaid internships

A short unpaid research internship with a genuine supervisor can be worth it. An unpaid “internship” where you do production work for a company with revenue is not an internship; it is unpaid labour with a certificate attached.

The test: are you being taught, or are you being used? If nobody senior is reviewing your work weekly, you are not learning anything you could not have learned alone.

The uncomfortable summary

The students who get good internships are, overwhelmingly, the students who were already doing the work before anyone paid them to. They did not build a portfolio to get an internship. They built things because they wanted to, and the portfolio was a side effect.

If you are six months out, that is your plan. Pick one hard thing, do it properly, write it up publicly. It will do more than another forty applications.

---

We take research and engineering interns twice a year, and we read what you have built before we read your CV. See open roles.