CAREERS · BHUGAON, PUNE

People who can read a P&L and a control loop in the same afternoon.

We build AI that runs inside someone else's plant, on hardware they own, with the uplink possibly pulled out. That work is multidisciplinary before it is technical, and commercial before it is either.

We read what you wrote down as closely as what you shipped.

How we hire Where you would work
THE BAR

We hire for the second brain, not the résumé.

The most valuable knowledge in any plant sits in the head of the person who has run it longest, and it has never been written down because nobody had a way to do it that did not mean forms. We are the company that argues this is the real problem — so we hire the people who already work that way.

That means two things we actually check. How you record what you learn: whether a decision you made six months ago can be reconstructed by someone who was not in the room. And whether anyone else can use it: notes that only make sense to their author are the same as no notes.

The rest follows from where the work happens. Our systems run on a customer's hardware, inside their boundary, often with no route out. You cannot debug that from a laptop in a café, and you cannot design it without understanding what the machine is physically doing. So the people who do well here are willing to stand next to the machine, and willing to be wrong in public about what they thought it was doing.

The combinations we look for

Mechanical + ML Controls + product Embedded + systems Physics + software Finance + code Design + field work Operations + data

None of these is a job title. They are the pairs that keep turning out to matter, because the interesting problems sit between two fields and the people who can hold both are rare. A single deep discipline is welcome — provided you are genuinely curious about the second one.

What your first year looks like

MONTH 01
On a shop floor
Not a site visit. You watch the process you are about to write software for, with the person who runs it, until you can describe it without our vocabulary.
QUARTER 02
A customer-facing number
Something a customer measures and would notice if it moved the wrong way. It is yours, including the part where you explain it when it does.
ONGOING
The commercial case for your own work
Builders here are expected to say what a thing costs, who pays for it and who would say no. Engineering that cannot be defended commercially does not ship.
WHERE YOU WOULD WORK

Three divisions, one core.

We do not run a separate research group. Everyone works on something a customer has agreed to pay for, and the platform is built by the people who have to deploy it. Pick the surface that interests you — the reading below is the real product page, not a job description.

DIVISION 01

VarahiField

Perception, control and diagnostics on the line, in the vehicle, on the plant floor. Small models on hardware the customer already owns, specified to run with the uplink pulled out.

Suits embedded and controls engineers, computer-vision people, and anyone who has debugged something with a multimeter in their other hand.
Read the division →
DIVISION 02

VarahiOne

The whole company on one air-gapped AI core — Workspace, Happen and VarahiConverse, plus the runtime underneath them: model serving, retrieval, agent scheduling, budgets and an append-only audit log.

Suits product engineers, distributed-systems people, and anyone who has run inference somewhere with no internet and no excuses.
Read the division →
DIVISION 03

AI Transformation

Making an enterprise AI-capable on its own hardware, then handing it over: scope and evaluation set, build, pilot, runbooks, and a second brain that stays behind when we leave.

Suits people who can sit with a plant head in the morning and an architecture in the afternoon, and who write clearly enough that the handover is real.
Read the division →
HOW WE HIRE

Three steps. No puzzles, no trick questions.

The whole process is designed to look like the job, because that is the only honest way for both sides to find out. You will meet the people you would actually work with, and you may use whatever tools you use normally — including AI, which we would think it odd of you to avoid.

01

A problem, not a puzzle

A real constraint from real work — usually one we hit in the last year. Bring your own tools and models; we care how you scope it, what you refuse to assume, and which question you ask first.

Roughly a conversation, not a take-home.

02

A week with the second brain

Paid. You work inside our memory and add to it — the same notes, the same audit trail, the same people. We read what you wrote down as closely as what you shipped.

Remote is fine for this step if Pune is not.

03

A business case, defended

Cost, risk, who says no and why. Fifteen minutes, in front of people who will disagree with you. We are not testing polish — we are testing whether you change your mind for a good reason.

You get the decision in writing, either way.

Step 02 is paid whether or not we make an offer. Asking for a week of unpaid work is a way of hiring only people who can afford to give it away.
OPEN ROLES

Four seats, and a standing door.

Every role below is on-site in Bhugaon, because the work needs you within reach of a shop floor. Apply to the one that fits, or use the general application if none of them do — we read those properly, and several people here arrived that way.

Embedded ML engineer

VarahiField
BHUGAON, PUNE · ON-SITE · FULL-TIME

Models that run on the customer's own hardware, next to the machine, with the uplink pulled out.

  • Small models on constrained hardware — quantisation, latency budgets, what fits in the box.
  • Reading bus signals and service history into something a technician will actually act on.
  • You will be on a shop floor often enough that steel-toe boots stop feeling like a costume.
Apply for this role

Computer vision engineer

VarahiField
BHUGAON, PUNE · ON-SITE · FULL-TIME

One vision core watching two things: whether the part is in tolerance, and whether anyone is working in a way that will injure them.

  • Inspection and pose estimation on live lines, measured as joint angles — never faces.
  • Camera placement, lighting and the unglamorous half of vision that decides whether it works.
  • Owning the false-positive rate in front of the plant head who has to live with it.
Apply for this role

Product engineer, sovereign runtime

VarahiOne
BHUGAON, PUNE · ON-SITE · FULL-TIME

The core underneath Workspace, Happen and VarahiConverse: model serving, retrieval, agent scheduling, budgets, and an append-only audit log.

  • Running inference somewhere with no internet, no vendor console and no excuses.
  • One index and one audit log across three modules that used to be separate products.
  • Install and upgrade paths for a box sitting inside someone else's network.
Apply for this role

AI transformation engineer

AI Transformation
BHUGAON, PUNE · ON-SITE · FULL-TIME

Making an enterprise AI-capable on its own hardware, then handing it over and leaving.

  • Scope and the evaluation set: sitting with domain experts until 'good' is written down.
  • Architecture the customer's own team can run after we go, and runbooks that say how.
  • A plant head in the morning, an architecture in the afternoon, both in plain words.
Apply for this role

None of these, but you think we are wrong about that

Tell us what you would build here and why it should exist. A good general application has changed what we were hiring for more than once.

General application
BEFORE YOU APPLY

The AI-First Builder Series runs in public.

Sessions, notes and code are open on GitHub — deep reads of real agent codebases, not talks. It is the fastest way for us to meet you, and the fastest way for you to find out whether you want to work like this.

Turning up, reading the code and disagreeing with us in the thread is a stronger application than a cover letter. Several people here arrived that way.

WHAT WE NOTICE
  • A question that shows you read past the README.
  • A correction to something we got wrong, with the reasoning attached.
  • Notes you kept in public, on anything, that someone else has used.
  • Work you finished when it stopped being interesting.
APPLY

Tell us what you have built, and what it cost.

No cover letter. The trade-off question below is the one we actually read first — a straight answer to it counts for more than anything else on the form.

What happens next

  • You get a reply either way, in writing.
  • If there is a fit, step one is a conversation about a real constraint — not a take-home.
  • Step two is a paid week inside our second brain, whether or not an offer follows.
305, 3rd Floor, Delta Square,
Siddhivinayak Vihar, Bhugaon,
Pune, Maharashtra 412115, India

Open 10:00 to 19:00 IST.

PDF or Word, up to 5 MB.

The role you are applying for, or what you would want to do here.

We keep what you send to consider you for this and future roles, and for nothing else.

Read the work before you write.

The division pages are the honest version of what you would be doing all day. They are worth ten minutes before you fill in the form above.

How we think about the work → Where this leads · 2030 → ← Back to the main site