Reviewed on 9 October 2026. An AI research scientist creates new knowledge: new methods, new evaluations or a new understanding of why models behave as they do, usually shared through peer-reviewed papers. That is a different job from building AI products, and the route into it is different too. This guide sets out the realistic path in India, the official programmes that fund or host early research careers, and a 12-month plan for building the evidence research groups look for.
Research scientist or ML engineer?
| AI research scientist | Machine learning engineer | |
|---|---|---|
| Main output | New methods, findings and papers | Working models and systems in production |
| Typical question | "Is there a better way to do this, and can we show it?" | "How do we make this work reliably for users?" |
| Usual qualification | Often a PhD, or substantial research experience | A bachelor's or master's degree with strong engineering skills |
| Evidence that counts | Publications, reproducible experiments, a research statement | Shipped systems, code quality, operational skills |
Many people move between the two roles, and industry labs hire both. Decide early which one you're aiming for, because the evidence you need to build is different.
Keep preparing
Continue with Sarkari Resume Templates₹299 — coaching के एक महीने से काफ़ी सस्ता / far cheaper than a month of coachingThe usual path in India
1. A strong foundation
Mathematics (linear algebra, probability, optimisation), programming, and core machine learning matter more than any single framework. A bachelor's degree in computer science, electrical engineering, mathematics or statistics is the common starting point, but researchers come from other quantitative fields too.
2. Graduate study, usually a PhD
Most research-scientist roles expect a PhD, or equivalent research output. In India, admission to M.Tech, MS (by research) and PhD programmes at many institutes uses GATE scores along with the institute's own selection. The GATE 2026 papers, organised by IIT Guwahati, included a Data Science & Artificial Intelligence (DA) paper alongside Computer Science (CS). Check each institute's admission notice to see which papers it accepts.
3. Funding: the Prime Minister's Research Fellowship
The PMRF guidelines document dated 14 January 2025 sets out direct-entry eligibility for PhD students at PMRF-granting institutes. Two of its criteria:
- A CGPA of at least 8.0 (on a 10-point scale) from a qualifying programme at IISc, the IITs, NITs, IISERs, IIEST or centrally funded IIITs, with selection to the PhD programme at a granting institute.
- For other recognised institutes, a CGPA of at least 8 plus a GATE score of at least 650 in the relevant paper, or a UGC/CSIR JRF rank of 100 or better (an NBHM fellowship also qualifies).
The document's pay matrix lists ₹70,000 for years 1 and 2 of the PhD, ₹75,000 for year 3 and ₹80,000 for years 4 and 5, which participating institutes describe as monthly fellowship amounts. It also lists a research grant of ₹2 lakh a year. Check pmrf.in and your institute's notice for the current cycle before applying.
4. Research roles before or instead of a PhD
Some industry labs hire recent graduates into research roles. Microsoft Research India's Research Fellows programme is one example. For the 2026 cycle, it asked for a BS/BE/BTech or MS/ME/MTech in computer science or related areas, and fellows spend one to two years at the lab's Bengaluru office. PhD students and PhD holders aren't eligible. The 2026 cycle ran from 7 January to 15 February 2026, with fellows starting in July 2026, so watch the page for the next cycle. A role like this lets you test whether research suits you before you commit to a PhD.
5. Where AI research happens in India
Industry labs include Microsoft Research India in Bengaluru, IBM Research – India, which says it was established in 1998 and has sites in Bengaluru and Gurgaon, and the AI lab Google announced in Bengaluru in 2019. Academic groups at IISc, the IITs and the IIITs also run active AI research. Read a group's recent papers before you contact it.
The government is also investing in AI capacity. A PIB release of 7 March 2024 records Cabinet approval of the IndiaAI Mission, with an outlay of ₹10,371.92 crore. Its FutureSkills pillar aims to increase AI courses at undergraduate, master's and PhD levels.
What research groups look for
- Research experience you can explain. A project where you formed a hypothesis, designed experiments, reported results honestly (including what didn't work) and can defend your choices.
- Writing. A clear technical report or paper. Peer-reviewed venues such as NeurIPS publish their calls for papers and workshop tracks. Workshops are a realistic first target.
- Reproducible code. A public repository that someone else can run to get your numbers.
- Recommendations from researchers who have supervised your work. PhD and fellowship applications usually ask for them.
A 12-month plan for a final-year student
The plan below is an illustrative schedule. Adjust the months to your own academic calendar.
| Months | Goal | Output |
|---|---|---|
| 1–2 | Choose a narrow area, such as evaluation of Indian-language models, and read 15–20 recent papers | A one-page summary of open problems, in your own words |
| 3–5 | Reproduce one recent paper's main result | A repository and a short report comparing your numbers with the paper's |
| 6–8 | Extend it with one new experiment, ablation or dataset | A results table with error bars or multiple random seeds |
| 9–10 | Write it up as a workshop-length paper and ask a faculty member for feedback | A draft with a clear claim, method, results and limitations |
| 11–12 | Apply to PhD programmes, PMRF where eligible, and research-fellow roles | A research statement built around your project |
Checklist for a reproduction project
- The paper's claim is written in one sentence before you start
- The dataset, its version and its licence are recorded
- Random seeds are fixed, and results come from more than one seed
- Your numbers sit next to the paper's, and any gap is explained
- Compute used (hardware and hours) is noted
- A README lets a stranger rerun everything
Common mistakes
- Collecting certificates instead of results. Short online courses don't substitute for a research project with a write-up.
- Chasing leaderboards without understanding. A small, well-explained finding beats an unexplained jump in score.
- Emailing professors generically. Mention a specific paper of theirs and say how your project relates to it.
- Ignoring the PhD question. If you want to lead research, plan for a PhD or an equivalent research track. Very few research-scientist roles skip it entirely.
The path is long, but it's well-defined. Build one honest, reproducible piece of research, get feedback on it, and use it to open the next door, whether that is a fellowship, a PhD place or a research-fellow role.