transPEAKtation

When a stadium empties out, every navigation app hands thousands of people the same "fastest" route, and that route stops being fast. transPEAKtation predicts where congestion will form around major events and distributes travelers across routes so the whole crowd gets from A to B faster. My role sat between product and engineering: I helped turn an open-ended traffic problem into a focused MVP, then built much of the data and backend layer that made it real.

what I did

1. 01AI helped define the product thesis and scope. Navigation tools optimize one traveler at a time, and we wanted to show that event-aware routing could improve outcomes across the network. For a weekend build, that meant prioritizing future travel-time selection, congestion context, and route explanations over everything else.
2. 02AI shaped the core flow: enter an origin, destination, and travel time → compare routes → see which upcoming events, incidents, and congestion affect each option. That set the product decisions downstream: event categories, route-card messaging, congestion overlays, time-aware filters, and search behavior.
3. 03AI broke the system into workstreams (Python ingestion workers, a FastAPI backend, routing/prediction logic, shared data contracts, and a Next.js frontend), defined how they talked to each other, and kept four people's parallel work converging into one working demo.
4. 04AI built the ingestion pipeline that pulls event and road data from DataSF Special Events, Caltrans, and CHP, then normalizes inconsistent schemas, deduplicates overlapping reports, geocodes locations, and writes to MongoDB Atlas and Tiger Data. It processed and stored 17,000+ road incident records, served through /events, /road-conditions, /traffic, and /plan.
5. 05AI kept ingestion off the request path with a dedicated POST /ingest/refresh endpoint, so fresh city data could be pulled and reprocessed without touching the main API.
6. 06AI made the system time-aware: events and incidents are filtered against when you plan to travel, not just what's happening now. That meant getting San Francisco local time right across services, modeling arrival and departure windows, and surfacing the relevant events next to each route.
7. 07AI integrated Mapbox Directions with OSRM and Nominatim fallbacks, worked through location search, route generation, networking, and environment config, and built UI pieces including incident markers, congestion visualization, route cards, and light/dark mode.
8. 08ABeyond the MVP, I explored where the product could go: community event submissions, organizer tools, boosted listings, and ticket-affiliate revenue. I also shaped the pitch around system-level impact, and the project won 2nd Place for Best Use of AWS and Best Use of Tiger Data at ShellHacks 2026.

built with