Careers
Seven roles across Roorkee and Mohali. We hire slowly and expect you to argue with us in the interview.
A place where you can say the model does not work
In most commercial data science, a scientist who reports that the model failed to beat the baseline has created a problem for somebody's quarter.
Here that is the job. The person who catches leakage before deployment has done more for the client than the person who squeezed out another point of accuracy, and is credited accordingly. Every scientist has the right to refuse to ship a model they cannot defend, and no one has ever been overruled.
You will present your own findings to clients, including the unwelcome ones. If you want your work filtered through a manager before it reaches the person paying for it, this is not the right firm.
What we actually offer
Time to read
Four hours a week, protected, for reading papers. Unused time is a management failure, not a saving.
Permission to say no
Every scientist here can refuse to ship a model they cannot defend. Nobody has ever been overruled.
Compute that is not rationed
You will not queue three days for a GPU to run an ablation.
You present your own work
To the client, with the limitations included. No manager sanitises your findings.
Publication is encouraged
If the work is novel and the client agrees, publish it. Your name, not the firm's.
Remote where the role allows
Two roles are remote friendly. We say which in the listing, not after you accept.
Seven positions
Own a problem from framing to deployment. You will be asked to justify your metric before your model.
Apply for this roleTake research code and make it something that runs at three in the morning without waking anyone.
Apply for this rolePipelines, warehouses, and the unglamorous work that makes every model above it possible.
Apply for this roleRetrieval, evaluation, and knowing when a smaller model is the correct answer.
Apply for this roleDetection and segmentation on real industrial imagery, not benchmark datasets.
Apply for this roleFeature stores, model registries, drift monitoring, and reproducible training runs.
Apply for this roleTranslate a vague business question into a measurable one. The hardest job on this list.
Apply for this roleNo roles open in this team right now.
Send us your work
A resume is fine. A notebook, a paper, or something you built is better.