Sumanas Sarma

Chief Technology Officer @Captur

London, GB

Impact: Productionised on-device AI (OTA models, CI/CD, Datadog), scoped/prioritised the Gen-2 shared-encoder rollout (~6 MB,~30 ms inference), and scaled the SDK to hundreds of cities.SDK platform & releases: Replaced ad-hoc local builds with CI/CD (GitHub Actions), formalised release gates, and enabled over-the-air model updates for iOS (Core ML) and Android (TFLite).Reliability & observability: Built hardware-in-the-loop soak tests (Raspberry-Pi rig + iOS/Android); instrumented Datadog RUM for crash/telemetry; filtered irrelevant session ingestion >75% and accelerated issue triage.Gen-2 architecture (leadership): Scoped and prioritised the shared-encoder + multi-head direction (MobileNetV3), owned the release plan, monitoring gates, and test strategy; the ML engineer delivered implementation. Results:~6 MB bundle,~30 ms iOS inference with Android parity; faster, smaller, more consistent rollout.Productisation & client enablement: Shipped React Native SDK in ~7 weeks; tuned image-quality policy per vertical (delivery vs micromobility); coordinated phased rollouts and success metrics with partners.Org & governance: Hired MLOps and test engineering, set vendor strategy (W&B, Datadog, Grafana), instituted semantic versioning/rollback, incident response, and risk-based decision reviews.Impact: Turned R&D into a reliable on-device platform deployed across hundreds of cities, handling thousands of verifications daily with measurable reliability and speed.