A paid AI internship where the code ships
The DBERT Fellowship places you on a real client codebase under senior review — production systems for incubated startups, not practice projects. You leave with merged commits, a performance stipend, and an experience letter a recruiter can verify.
What the fellowship gives you
Real client codebases
Write secure API routes, configure databases, and deploy model pipelines for incubated startups.
Performance stipends
₹5,000 to ₹18,000, paid against delivery milestones and code quality rather than hours logged.
Verifiable letters
Experience letters and LORs signed by the DBERT engineers who reviewed your pull requests.
What a paid AI internship in India usually gets wrong
Search for a paid AI internship in India and most of what comes back is a course with an internship label stapled to it. The work is a sandbox: a to-do app, a chatbot against a public API, a notebook that runs once and is never opened again. The certificate at the end certifies attendance. Recruiters have learned to read straight past it, which is why so many candidates with three internships on a CV still cannot answer a question about production incidents, migrations, or code review.
The gap is not knowledge. It is evidence. An engineer who has shipped to a repository with users on the other end has a different set of habits — they write migrations that can be rolled back, they read a stack trace before they read Stack Overflow, and they can explain why a pull request was rejected. Those habits are visible in an interview within about ten minutes, and no amount of coursework simulates them. For deeper strategic guidance on separating legitimate technical engagements from marketing sandboxes, review our 2026 Guide to Paid AI Internships in India.
What the DBERT Fellowship actually is
The fellowship is a three-month, selection-based programme that places you inside a development squad building software for DBERT's own incubated startups and enterprise clients. It sits at the top of the AI career programs ladder, after the Accelerate engineering programme, and it is the rung where the work stops being ours and starts being a client's.
That distinction drives everything else about the programme. Because a client is waiting on the output, your pull requests are reviewed by senior engineers against the same bar as any other contributor's, and rejected on the same grounds. Because the systems are live, the work is unglamorous in the specific way real work is unglamorous — schema changes, retrieval quality, error handling, the long tail of things that only matter when someone is actually using the software.
The work itself
Squads are assigned to a system rather than a curriculum, so the exact stack depends on the client. Across recent squads that has meant:
- REST and token-streaming API routes, including auth and rate limiting
- Postgres schema design and migrations, plus vector-index configuration for retrieval
- Retrieval-augmented generation pipelines — chunking, embedding, evaluation of answer quality
- Front-end interfaces against those APIs, built to the client's design
- Deployment, environment configuration, and the observability needed to debug it later
You will not touch all of it. Squads specialise, and a fellow who spends three months getting a retrieval pipeline genuinely good has a better story than one who touched everything shallowly. Explore how these production vector components come together in our technical workshop on Production RAG Architecture from Scratch.
Selection: two ways in
The first route is completing DBERT Accelerate, which is how most fellows arrive — by then we have seen months of their code and know what they can be trusted with. The second is the external code screening challenge, open to anyone regardless of where they learned. The screening exists so that self-taught engineers and graduates of other programmes are not locked out of the pipeline by not having bought our course first.
Either way, the application asks for code — a repository, a deployed project, a screening submission — rather than a list of coursework. That is what the decision is made on.
The stipend, stated plainly
The stipend ranges from ₹5,000 to ₹18,000 and is performance-based: it is paid against delivery milestones and code quality, not hours logged. Nobody starts at the top of the band. A fellow who ships reviewed, merged work across the full three months earns materially more than one who does not, and we would rather say that here than have you discover it in month two. To see how intermediate and advanced performance directly scales into full-time compensation across Indian technology hubs, inspect our analysis of 2026 AI Developer Salaries in India.
The programme fee is ₹1,599 for the three months, covering mentorship, code review, and certificate and letter issuance. Fees are reviewed each cohort, so treat that as the current figure rather than a permanent one.
What you leave with
Three things, in descending order of how much they matter in an interview:
- Merged commits on a system with users. Public where the client permits it, describable in detail where they do not. This is the part that changes how the interview goes.
- An experience letter naming the engineer who reviewed your work and the system you contributed to — not a certificate of attendance. It carries a certificate ID a recruiter can check independently at the DBERT certificate verification page without contacting us. A letter nobody can verify is worth what it costs to print.
- Visibility to partner startups. They review fellow performance logs when hiring, and the AI jobs board for verified builders is restricted to people with that record.
What we do not claim
We do not guarantee placement. Any programme in India promising a job at the end is either selecting so hard that the guarantee is meaningless or is going to disappoint someone. A referral is not an offer, and we will not describe it as one.
We do not publish a placement percentage or an acceptance rate. We would rather publish nothing than publish a number we cannot substantiate to a prospective fellow who asks how it was calculated. When the cohort history is long enough to state honestly, it will appear on the jobs page with its method attached.
We do not promise the top of the stipend band. It is performance-based and most fellows land in the middle of it.
How three months are structured
Technical onboarding and squad assignment
You are assigned to a development squad under an engineering mentor, given repository access, and set up locally against the real database schema. Week one is environment, codebase tour, and your first reviewed pull request.
Active contribution sprints
You design interfaces, configure vector indexes, build API routes, and take part in code review — on the client repository, not a fork. Every pull request is reviewed and graded by a senior engineer before it merges.
Evaluation and career handoff
Final module hardening, then the paperwork that makes the work legible to a recruiter: an experience letter tied to a verifiable certificate ID, your performance-based stipend, and introductions where a partner startup is hiring.
Why this is structured as a fellowship, not a course
A course can be run at any scale — the marginal cost of one more student is close to zero. Client work cannot. Every fellow consumes senior engineering review time on a repository with a deadline, which is the scarcest thing DBERT has. That constraint is why the programme is selective, why squads are small, and why we cannot simply admit everyone who applies.
It is also why the stipend is performance-based rather than flat. The programme only works if fellows ship: a squad carrying someone who does not is a squad missing a client deadline.
What makes the fellowship different
- Not sandbox toys — production databases and live API pipelines, on the client repository.
- Senior code review — every pull request reviewed and graded before it merges.
- Independently verifiable — the letter carries an ID a recruiter can check without asking us.
- Direct hiring visibility — partner startups read performance logs when they hire.
Paid AI internship — common questions
Yes, with a specific structure worth understanding before you apply. The stipend is performance-based and ranges from ₹5,000 to ₹18,000, paid against delivery milestones and code quality rather than hours logged. It is not a salary and it is not guaranteed at the top of the band — a fellow who ships reviewed, merged work across the full programme earns more than one who does not.
No. There are two routes in. The first is completing DBERT Accelerate, which is how most fellows arrive. The second is passing our external code screening challenge, which is open to anyone regardless of where they learned to code. The screening exists precisely so that self-taught engineers and graduates of other programmes are not locked out.
No, and you should be sceptical of any programme in India that does. What we actually provide is a verifiable record: an experience letter signed by the engineers who reviewed your code, a certificate ID anyone can check at dbert.online/verify, and commits on a real client repository. Partner startups review fellow performance logs when they hire, so the pipeline is real, but a referral is not an offer and we will not describe it as one.
Production work on systems that have users. That has meant REST and streaming API routes, database and vector-index configuration, retrieval pipelines, front-end interfaces, and deployment plumbing for incubated startups and enterprise clients. It does not mean tutorial rebuilds, sandbox exercises, or a capstone that nobody runs after you leave.
The programme runs three months and the fee is ₹1,599. The fee covers mentorship, code review, and the certification and letter issuance at the end; the stipend is separate and flows to you. Fees are reviewed each cohort, so treat the figure on this page as current rather than permanent.
It is remote and squad-based. Sprints are structured around delivery milestones rather than fixed office hours, which is what makes it workable alongside a final-year course load. It is still real client work with real deadlines, so it is not a zero-effort add-on — fellows who treat it as one tend to finish at the bottom of the stipend band.
Two ways. It names the engineer who reviewed your work and the system you contributed to, rather than certifying attendance; and it carries a certificate ID that a recruiter can verify independently at dbert.online/verify without contacting us. A letter nobody can check is worth roughly what it costs to print.
Applications run through the DBERT internship platform at internship.dbert.online. The application asks for your code — a repository, a deployed project, or your screening submission — rather than a CV of coursework, because that is what the selection is actually based on.
Applications open for the next cohort
Entry is by completing DBERT Accelerate or by passing the external code screening challenge. Bring code, not a CV.
Apply for the paid AI fellowship