AI Engineer
Make models useful in production, where they are judged on what they return rather than on a benchmark.
The work.
You would work across Helium One and ModelBeat: retrieval that has to be right, routing that has to be cheap, and evaluation that has to be honest about what is failing.
This is not research. The work is the distance between a model that scores well and a model a team trusts on a Tuesday afternoon.
Small team. What you build ships in days, and you own it after it does.
What we would hope to see.
- You have put something model-shaped in front of real users and watched it break
- You can tell the difference between a prompt problem, a retrieval problem and a model problem
- You write code other people have to maintain
Pune · Full time · Engineering
Apply.
Ten minutes, honestly. Three questions we wrote for this role, three we ask everyone, and no cover letter.
Not this one? Back to the open roles.
