🚆 CalTrain AI
A real-time voice and chat assistant for Caltrain schedules powered by OpenAI Realtime API, WebRTC, and official GTFS data, delivering live departures, delays, and fares in all languages.
External Links
Achievements
Built and shipped CalTrain AI as a real-time voice assistant for Caltrain schedules during the Supabase AI Meetup in Palo Alto, integrating OpenAI Realtime API, GTFS static and realtime data, and Vercel Edge functions.
Presented at Supabase Meetup
Delivered a talk on building production-ready realtime AI agents using Supabase, covering real-time streaming, memory storage, multi-agent architectures, low-latency pipelines, and hallucination guardrails—showcasing how Supabase enables rapid prototyping to production with minimal backend work.
𝗦𝗶𝘁𝘂𝗮𝘁𝗶𝗼𝗻
Bay Area Caltrain riders often rely on static PDFs or clunky schedule tools, making it hard to quickly check the next train, delays, or fares—especially on mobile or while on the move.
𝗧𝗮𝘀𝗸
Create a real-time voice and chat assistant that understands natural language, supports English and Spanish, and can answer questions about Caltrain schedules, fares, and directions using official GTFS data.
𝗔𝗰𝘁𝗶𝗼𝗻
Built a Next.js app with a dual-mode interface for voice (WebRTC + OpenAI Realtime API) and chat (GPT-5.1-nano) using optimized low-temperature prompts for concise, factual answers.
Integrated Caltrain GTFS static and GTFS Realtime feeds with a caching pipeline (Vercel KV / file-based cache) to serve live departures, delays, and zone-based fares with low latency.
Implemented tools for directions, fare calculation, nearest-station lookup, and multilingual responses, with origin validation and edge runtime APIs for performance and security.
𝗥𝗲𝘀𝘂𝗹𝘁
CalTrain AI provides commuters with a fast, hands-free way to check the next train, delays, and fares, with responses streaming in real time and displayed in interactive schedule and fare cards on desktop and mobile.