ROUTE LINK

Inspiration

RouteLink came from a real problem my family faced while travelling. My mother accidentally left her handbag at a restaurant, and we only realized it after we had already driven far away. On the return trip, my father asked an employee who was near the restaurant to pick it up, but because both sides were near a highway and moving through traffic, deciding where to meet became surprisingly difficult. There were calls, location sharing, back-and-forth discussion, and uncertainty about what point would be fair and practical for both people.

That made me think: maps are great at taking one person from A to B, but they do not really help two people coordinate a quick mid-route meetup. RouteLink is built for that moment.

What it does

RouteLink helps two people find a practical meetup point based on their real routes, not just straight-line distance. Users can create an anonymous session, share an invite link, stream live GPS or enter test locations, and get road-aware meetup options with travel-time breakdowns.

The app suggests real nearby places instead of raw coordinates, shows who has to travel how much, flags unfair options, supports voting and confirmation, and then hands off the confirmed point to Google Maps.

How I built it

I built RouteLink with Next.js, TypeScript, Tailwind CSS, MapLibre, Mapbox APIs, Upstash Redis, Ably, and the OpenAI API. Mapbox provides the factual map, route, ETA, geocoding, and venue data. Redis stores anonymous expiring sessions, and Ably keeps the session synced in realtime.

GPT-5.6 Terra is used inside the product for two judgment-based tasks: explaining why a meetup option is good or unfair, and selecting the best venue from real Mapbox POIs. The model does not invent locations; it only reasons over verified route and venue data.

Challenges

The hardest part was making the routing feel realistic. A midpoint on a map is not always useful — roads, destinations, highways, traffic, and direction of travel all matter. I had to handle different cases like one person driving while the other is stationary, two people driving in the same direction, and fallback negotiation points when there is no clean shared route.

Another challenge was making live GPS and manual testing work together without fighting each other. I added clear source controls so a participant can use live tracking or manual test coordinates, but not both at the same time.

What I learned

I learned how important it is to separate facts from judgment in an AI app. RouteLink uses Mapbox for facts such as roads, ETAs, and real venues, then uses GPT-5.6 Terra only for human-like judgment: explaining tradeoffs and picking a suitable place from real options.

I also learned how much polish matters in a coordination app. Small details like a case badge, live/manual status, map legend, copy coordinates, and maps handoff make the product much easier to understand in a real stressful moment.

What I’m proud of

I’m proud that RouteLink started from a very ordinary family problem and became a working app that could genuinely help in similar situations. It is not just “find the midpoint”; it tries to answer the more human question: where can we meet that is practical, fair, and easy to act on?

Built With

• ably
• gps
• gpt
• mapbox
• maplibre
• maps
• next.js
• openai
• realtime
• redish
• routing
• tailwindcss
• typescript
• uptash
• vercel