-
-
Landing Page
-
Canvas View
Inspiration
As an AI product manager and software developer, I've been deeply immersed in building GUI-based agents—AI systems that don't just chat, but actually do things on interfaces. When I saw the Yelp API hackathon, it clicked: trip planning is the perfect use case for an agentic AI. Traditional trip planning is tedious—you search, compare, bookmark, forget, repeat. What if an AI could watch you describe what you want and just... build the plan for you, visually, in real-time? The free API access made it possible to experiment with combining Yelp's rich business data with conversational AI to create something that feels like having a local friend plan your day.
What it does
Yelp Yellow is an AI-powered visual trip planner. You describe what you're in the mood for—"a romantic night out" or "a family adventure day"—and watch as the AI interprets your intent, builds a dynamic timeline, and curates the perfect spots from Yelp for each stop.
The plan appears as an interactive canvas where you can:
- Watch it build — See each stop decided in real-time with AI reasoning
- Click any node — View the selected spot's details (photos, address, reviews)
- Tune with natural language — Type "something more casual" to get alternatives
- Drag to customize — Add stops from a palette, drop onto connections to insert mid-itinerary
- Export & share — Copy your finalized plan
How we built it
- Frontend: React with a custom canvas system for the visual timeline. Nodes are draggable, edges are interactive drop targets, and the whole thing streams updates via Server-Sent Events.
- Backend: Flask/Python serving a REST API that orchestrates between Yelp Fusion (business search) and Yelp AI API (intelligent curation and conversation).
- AI Pipeline: The system interprets prompts to dynamically generate appropriate timelines (night out ≠ full day), then iteratively curates each stop using the Yelp AI API with context about the overall trip.
- Infrastructure: Dockerized for easy deployment, with gunicorn serving the production app.
Challenges we ran into
- Streaming AI responses — Making the plan "build itself" visually required implementing SSE streaming and carefully managing React state as nodes arrived asynchronously.
- Intelligent timeline generation — Getting the AI to understand that "night on the town" shouldn't include breakfast took iterative prompt interpretation logic.
- Request cancellation — Users clicking around fast could cause stale API responses to overwrite fresh data. We implemented request ID tracking to ignore outdated responses.
- API timeout management — The Yelp AI API calls could take time; we had to tune gunicorn worker timeouts to prevent premature termination.
Accomplishments that we're proud of
- The "wow" moment — Watching the plan build itself node-by-node, with the AI explaining each pick, genuinely feels magical.
- Drop-on-edge insertion — Dragging a "Drinks" node onto a connection and having it splice into the itinerary with auto-populated suggestions is deeply satisfying.
- Conversational tuning that works — Typing "sushi instead" in the refinement box and seeing the suggestions actually change to sushi restaurants validates the whole agentic approach.
- Clean architecture — Despite rapid development, the codebase is modular: Yelp client, AI curator, and trip planner are cleanly separated.
What we learned
- Agentic UIs need feedback loops — Users need to see the AI "thinking" and making decisions, not just get a final result. The streaming build screen transformed the UX.
- Yelp's data is incredibly rich — Categories, attributes, photos, hours—there's so much to work with for intelligent filtering and presentation.
- Conversation context matters — Maintaining chat history per node allows genuinely iterative refinement, not just one-shot queries.
- Visual planning beats list-based planning — Seeing your day as a connected graph makes it tangible in a way bullet points never could.
What's next for Yelp Yellow
- Multi-day trips — Full vacation planning with day-by-day views and travel time between stops
- Collaborative planning — Share a live canvas with friends/family to plan together
- Calendar integration — Export to Google Calendar with reservation times
- Mobile companion — Day-of mode that guides you through your itinerary with real-time updates
- Smarter constraints — "We have a stroller" or "someone's vegetarian" propagating across all stops
- Yelp Waitlist/Reservations integration — Book directly from the plan
- Conversational AI API Optimization - As the API improves, performance of the tool will improve.
