Industries
Six sectors where we have made the expensive mistakes already, on somebody else's dataset.
Domain knowledge is not optional
A data scientist who has never seen a credit file will build a default model that quietly encodes the collections flag and reports a spectacular result.
Knowing what a column means, and when it was written, is not a soft skill. It is the difference between a model and a leak. These are the six sectors where we already know which columns lie.
Manufacturing and Industrial
Demand forecasting, predictive maintenance, and visual quality inspection under real conditions.
Factory lighting, dust, and vibration destroy models trained on clean benchmark imagery. We test on your floor.
Financial Services
Credit risk, fraud detection, and document extraction, in a sector where a model must be explainable to a regulator.
A model that cannot explain a rejection is not deployable in lending, whatever its accuracy.
Healthcare and Life Sciences
Clinical decision support, imaging triage, and operational forecasting, always with a human in the loop.
We build systems that assist a clinician. We do not build systems that replace one.
Logistics and Supply Chain
Routing, ETA prediction, and demand sensing where the cost of being wrong is measurable to the rupee.
Forecast accuracy matters less than knowing when the forecast should not be trusted.
Retail and Ecommerce
Pricing, recommendation, and inventory, evaluated against a holdout of real customers rather than an offline metric.
Offline recommendation metrics correlate poorly with revenue. We insist on an online test.
Public Sector and Research
Analysis, evaluation, and independent assurance for institutions accountable to the public.
Work funded by the public should be reproducible by the public. We build for that.
Where we decline
We do not build autonomous trading systems, safety critical control loops for vehicles or medical devices, or any model that makes a consequential decision about a person without human review.
The first two demand a specialisation and a safety culture we do not have. The third we decline on principle, and we would rather lose the work than argue about it later.
Four questions before we accept
Send us your data problem. We will tell you whether it is one.
A working data scientist reads every enquiry. You will get an honest read on whether your data can support the model you have in mind, before anyone quotes you a number.