Insights
Field notes from production deployments.
Engineering posts on accuracy SLAs, edge inference, DPDP compliance, and what we have learned shipping safety vision into Indian schools, hospitals, and plants.
-
Eval-first is the only honest way to ship AI
The SLA argument generalises. Whether you're shipping safety vision, an agentic feature, a VFX pipeline, or a research model, the single highest-leverage choice is defining "good" before you train.
eval production ai responsible ai -
Why per-module accuracy SLA is the only honest way to sell AI safety
A single "99% accurate" headline is a marketing artefact. Real safety deployments have to be measured per module, per site, monthly — and the contract has to mean it.
sla computer vision responsible ai