trequila
liveA full-stack AI travel planner that turns a single prompt into a complete trip itinerary — places, weather, photos, all streamed in real time.
- Distributed, event-driven Go backend on NATS JetStream, fanning each HTTP request across 4 parallel micro-workers — cut end-to-end latency from ~2.5 min to 50–60s.
- Streamed partial itineraries to the Next.js frontend via Server-Sent Events, putting first results on screen in under 8s.
- Ranked spots with a linear ML model trained offline over 50 destinations, then sequenced daily routes to minimize travel distance via a 2-opt greedy TSP solver.
- Managed five independent async state slices arriving out of order, driving a live stage-progress UI with per-stage counts instead of one loading spinner.