Generative AI Jobs in India: What Employers Are Actually Hiring For

Most writing about generative AI jobs in India is about the size of the market. It quotes a growth rate, a projected number of roles, a figure for what the technology will add to the economy, and leaves the reader with the impression that the hiring itself is straightforward. It is not. The market is real, and what employers are actually hiring for bears little resemblance to what the headlines describe.

This page sets out what a generative AI role in India tends to consist of, which roles exist in volume and which are still mostly aspiration, and how an employer in financial services or insurance should write a brief for one. It carries no market size figures, because the ones in circulation are rarely sourced.

What does a generative AI role in India actually involve?

Far more integration and evaluation than model building, and the brief should say so.

The mental image behind most generative AI job descriptions is a researcher training a model. Almost nobody in India is hired to do that, and the handful who are already have jobs. The work that exists in volume is applying models that already exist: connecting them to a company own data, building the retrieval and guardrails around them, evaluating whether the output is good enough to act on, and making the whole thing run reliably in production.

That is engineering work with a statistical flavour rather than research, and it draws on the same disciplines the data profession already has. Someone who can build a reliable pipeline, reason about data quality and explain a limitation to a non specialist is most of the way to being useful on a generative AI team. Someone who has only ever fine tuned a model in a notebook is not. The distinction between the three core data roles is set out in our note on data engineer against data scientist against data analyst, and it holds here.

Which roles exist in volume, and which are mostly titles?

Application and platform roles exist. Pure research roles are rare. “Prompt engineer” as a standalone job has largely not survived contact with reality.

The practical lesson for a hiring manager is that the title on the advert should describe the row the work actually sits in. Advertising an applied scientist role and filling it with pipeline work is the most reliable way to lose the hire inside a year, and it is the most common error in this market.

Why is the talent gap real and the graduate surplus also real?

Because the gap is in production experience, which no course supplies and which most graduates have not had the chance to acquire.

India produces an enormous number of people who have completed a course in machine learning or generative AI, and a much smaller number who have shipped something that ran for a year and broke occasionally. Employers are hiring for the second thing. The gap between the two is the same employability gap that runs through the wider Indian graduate market, set out in our piece on why a graduate surplus still leaves roles unfilled, and it is at its most acute in this field because the technology is new enough that few people have had time to accumulate the experience.

This has a direct consequence for how employers should hire. Screening on certificates and course completions produces a large pool of people who look similar and cannot be told apart. Screening on evidence of a working system, even a small one, produces a much smaller pool of people who can. The second pool is the one worth interviewing.

What is different for financial services employers?

The tolerance for a wrong answer is far lower, and the regulator will ask how the output was controlled.

A generative AI system that occasionally produces a confident wrong answer is an irritation in most industries and a regulatory event in this one. A model that drafts a customer communication, summarises a claim or supports a credit decision has to be evaluated, monitored and explainable to a supervisor in a way that a marketing copy generator does not. That changes the hire.

Financial services teams need people who understand model risk as a discipline, who can design evaluation that would satisfy an internal audit function, and who treat a control as part of the product rather than as a constraint on it. That profile overlaps with model validation and quantitative risk more than with consumer technology, and employers who brief for a consumer technology profile end up with someone who is fast, capable and uncomfortable with the constraints. Brief for the risk aware version from the start.

How should an employer write the brief?

Around the system the person will own and the state of the data underneath it, not around a list of frameworks.

• Name the system. What it does, who uses the output, and what happens when it is wrong. That sentence tells a good candidate more than a page of requirements.

• Describe the data honestly. Candidates who join a messy environment knowingly stay; candidates who discover it in week three leave.

• Separate what must be present from what can be learned. Most generative AI tooling is learnable in weeks by someone with the underlying engineering.

• State whether the role builds, evaluates or operates. Those are different jobs and candidates self select accurately when told.

• Say what controls exist. In financial services that is a selling point to the right candidate and a warning to the wrong one.

Employers building these teams inside a capability centre should also decide early where the team sits and how it is kept, since the people who can do this work are approached constantly. Our notes on choosing a location for specialist work and keeping senior specialists once hired cover both.

What should a candidate do to be hireable?

Build one thing that runs, document it, and be able to say what went wrong.

The candidates who get through in this market are rarely the ones with the longest list of courses. They are the ones who can walk an interviewer through a system they built, however small, and describe the point at which its output was wrong and what they did about it. That story demonstrates the evaluation habit employers are screening for, and it cannot be faked from a certificate. A candidate with one such project and a clear account of it outranks one with five completions and no working system.

Is the market cooling or still growing?

Growing in substance and cooling in noise, which is healthier than it sounds.

The period in which every company announced a generative AI initiative and hired for it without a clear use has passed. What remains is a smaller number of employers with a working system and a real need to staff it, and that demand is steadier and better paid than the earlier wave. For candidates it means the premium has moved from being early to being able to demonstrate something. For employers it means the pool of people with genuine production experience is still small, still well known, and still being courted, and that a brief written for the noise period will not attract them.