We build AI systems for large industrial customers — satellite imagery pipelines, knowledge systems and agent platforms. When a user hits a problem, somebody has to work out what actually happened and get it to the right person.
That is this job. It is a first year, and it is a real one.
What we build
Varahi is an engineering company in Pune. We build AI that runs on hardware the customer owns, inside their own building, and keeps working when the internet does not. There are three product lines, and this is all of them:
For the companies that design, build and service machines. VEDA reads a machine's own signals and helps a technician find a fault. Nirdesh replaces the 400-page manual with something you can ask out loud, hands dirty. VCAD turns a description of a part into geometry you can manufacture.
One air-gapped AI core a small or mid-sized company runs everything on: Workspace for mail and files, Happen for CRM and outreach, VarahiConverse for projects and people. One box, one record, no vendor holding a slice of it.
Sold as engineering, not advice. We make an enterprise AI-capable on its own hardware, then hand it over and leave — architecture, evaluation set, pilot, runbooks.
One idea sits under all three. A company already knows how its machines fail and how its work gets done. That knowledge lives in people's heads and it leaves when they do. Our job is to write it down as a by-product of the work, so it stays in the building.
The work, honestly
About half of it is first-line support. Something comes in and you decide what it is: a question, a data problem, or a real bug. Then you write it up clearly enough that the person who fixes it does not have to come back with three questions.
About a quarter is working with the AI team — reviewing what a model produced, labelling data, running test cases, and taking deployment steps under supervision.
The last quarter is documentation.
Some of it is repetitive, and we are not going to pretend otherwise. What you get in return is that you will see every way these systems fail. That is why people who spend a year doing this become better engineers than people who spent the same year writing code nobody deployed.
You will see every way these systems fail. That is the point of the year.
Where it leads
At twelve to eighteen months you move into QA automation, business analysis, MLOps or data quality. Which one is your choice. At six months we will tell you honestly which we think fits.
What we need
- A degree in engineering, computer science or IT.
- Clear written English. Half this job is writing something down so somebody else can act on it.
- Care with detail.
- Python, SQL or Git help, but are not required.
No experience needed. We will train you.
Not for you if
You want to write model code from month one. That job exists here. This is not it, and we would rather say so now than in your third week.
How to apply
Send a CV and a short note answering one question: describe something that did not work, and how you worked out why. Anything at all — a project, a laptop, a bicycle.
No cover letter.
Apply by email → or use the form on our careers page →
You get a reply either way, in writing.
305, 3rd Floor, Delta Square, Siddhivinayak Vihar, Bhugaon, Pune, Maharashtra 412115 — near Chandani Chowk, on the road out to Pirangut.
This is one of the seats open at our Bhugaon office, near Chandani Chowk on the way to Pirangut. The careers page has the rest, plus how our hiring process actually runs.
All open roles →