Detour

Detour gives you AI travel guides with real personalities. Build a custom itinerary, follow walking routes, and unlock fun facts as you explore places that actually match your vibe.

Comment

Inspiration

Travel apps give you lists. Generic "top 10" recommendations that feel the same everywhere you go. We wanted something different--something that felt like exploring with a friend who actually knows the city. What if you could bring a local historian, a foodie, an adventure scout, and more on every trip? What if they could team up to find the perfect spots just for you?

That's Detour. Your AI travel crew, ready to explore.

What it does

Detour lets you pick from 9 distinct AI guide personalities--or combine multiple guides for hybrid recommendations:

  • 🏛️ Historian - Hidden stories & landmarks
  • 🍽️ Foodie - Authentic local flavors
  • 🧭 Adventurer - Thrills & nature pockets
  • 💻 Tech Guru - Innovation hubs & startups
  • 🏗️ Architect - Buildings & urban design
  • 🌙 Night Owl - Nightlife & late-night eats
  • 🎨 Artist - Murals & creative spaces
  • ✨ Local Oddball - Quirks & curiosities
  • 🏟️ Sports Buff - Stadiums & athletic history

Select one guide or mix several--like Foodie + Historian to find restaurants with fascinating backstories. The AI searches the web for real, verified fun facts about each place, then builds you an interactive walking journey.

Key features:

  • Real venue data powered by Foursquare Places API
  • Drag-to-reorder itinerary builder
  • Interactive map with numbered markers and route visualization
  • Journey mode with web-searched (powered by Tavily) fun facts for each stop

How we built it

  • React Native + Expo – Cross-platform mobile app
  • DigitalOcean AI Agent – Powers our 9 guide personalities with web search capabilities for real-time fact verification
  • Foursquare Places API – Fetches real, nearby venues based on location and guide type
  • OSRM (Open Source Routing Machine) – Calculates walking routes with accurate distance-based time estimates

Challenges we ran into

Getting the AI to actually search: Our guides would happily make up "fun facts" from training data. We had to craft prompts that force the AI to use its web search tool for every place, ensuring users get real, verified information--not AI hallucinations.

Multi-guide personality blending: Combining two guides (like Foodie + Architect) required careful prompt engineering to blend perspectives without creating a confused, generic response.

Accurate walking times: The OSRM public server only supports driving routes by default, so we had to calculate walking times ourselves based on realistic urban walking speeds (~4.3 km/h), accounting for crosswalks, stairs, and navigation.

Reasoning model quirks: The AI sometimes got "stuck thinking"--producing reasoning but no final output. This was fixed by careful model selection based on user inputs.

Accomplishments that we're proud of

  • 9 unique guide personalities that genuinely feel different to interact with
  • Multi-guide selection that creates novel, blended recommendations
  • Real facts, not hallucinations – The AI searches the web for every place it recommends
  • Polished, native UI with smooth animations and intuitive interactions
  • Actually useful walking times that reflect reality, not super human speed

What we learned

  • AI agents need very explicit instructions to use their tools consistently. Bigger models ≠ better.
  • Public APIs have annoying limitations (OSRM's walking profile) that require unique solutions.
  • How to use multiple agents together to meet user's needs.

What's next for Detour

  • Save & share itineraries – Export your journey or share with friends
  • Offline mode – Download journeys for areas with unstable connectivity
  • User-contributed guides – Let locals create & share their own guide personalities
Share this project:

Updates