Starting Phase 2 of Devtrails

Spent the last few weeks building GigGuard — a parametric insurance platform for Zomato and Swiggy delivery partners in India. The idea is simple: when it rains above 15mm/hr in your zone, you get paid automatically. No claim form. No waiting. Money in your UPI before you've dried off.
Phase 1 shipped a full working demo — Next.js frontend, Node.js backend, PostgreSQL, and a Python ML service for dynamic premium pricing and fraud detection. Phase 2 has been the interesting part: replacing text-based zone matching with Uber's H3 hexagonal grid for sub-kilometre payout precision, wiring a Thompson Sampling contextual bandit that learns which coverage tier each worker segment is most likely to buy, and running a SAC reinforcement learning agent in shadow mode to self-tune premiums against our loss ratio target.
The fraud detection problem turned out to be the hardest thing we tackled. GPS spoofing rings on Telegram are a real, organised threat to parametric insurance. Our answer was a Behavioral Coherence Score — four independent signal layers (accelerometer patterns, battery drain rate, cell tower triangulation, platform online status) that a spoofer has to defeat simultaneously to look legitimate. In practice they can't.
Built for Guidewire DEVTrails 2026. Still a lot of road ahead — GNN fraud detection, smart contracts on Polygon, causal inference for payout validation — but the core engine works and the tests are green.
If you're working on InsurTech, gig economy financial products, or parametric risk anywhere in South/Southeast Asia, I'd love to connect.