Practical case studies and engineering notes from the team — what we shipped, what we learned, and what we'd do differently.
Lessons from deploying an automatic number plate recognition system on Jetson devices at Indian toll gates — detection, OCR, validation, and the long tail of adversarial weather.
How we shipped a phone-only exercise recognition and rep-counting app that runs at 30 FPS without a wearable — and the pose-estimation tricks that made it work.
The agentic pipeline we built for SA Glass — multi-agent extraction, live catalogue validation, and a human-in-the-loop layer that makes 78% of orders fully automated.
A practical guide to behavior analytics on CCTV feeds — multi-object tracking, re-identification, dwell-time modeling, and how we keep precision above 90%.
Layout-aware document AI applied to one of the messiest PDFs in sport — bloodstock auction catalogues. How we turn 300-page booklets into searchable pedigree graphs.
Notes from the engineering side — what we build once and deploy many times: data pipelines, evaluation harnesses, and on-call playbooks for model services.