Taste Transplant

Moved cities? Keep your taste. An agent uses Qloo's taste graph to map the places in your new city that match what you love, with Qloo evidence behind every pick.

Comment

Click Try the Demo on the landing page for a one-click run (no typing or keys needed).

Inspiration

Moving to a new city, say from Pune to London, means losing the places, sounds and communities that fit you. Asking a generic AI gives popular-place lists and often just echoes your favorites back as venue names. We wanted an agent that actually understands taste.

What it does

TasteTransplant learns your taste from the things you love (artists, films, restaurants, brands, books), then builds a personal map of your destination city: where to eat, what to listen to, where to shop, what to watch and which neighborhoods fit your vibe. Every pick shows the Qloo evidence behind it (endpoint, signals and location filter), and a compare page shows the same request answered by a plain LLM with no Qloo, so the difference is visible.

Key features:

  • Taste onboarding with Qloo entity search and a one-click demo profile
  • A tool-calling agent that queries Qloo with your taste as signals and the destination as a location filter
  • Evidence drawer on every pick, plus an interactive map with Google Maps links
  • Side-by-side comparison against a plain LLM: [X] of 10 of its suggestions could not be matched to Qloo's taste results for the same profile
  • A concierge chat to refine results ("quieter", "cheaper") that re-queries Qloo
  • Saved plans with public share links

How Qloo powers it

Recommendations come from Qloo, not from the LLM:

  • /search resolves your favorites to real Qloo entities
  • /v2/tags resolves style and genre keywords into taste tags
  • /v2/insights returns taste-matched places, artists, films and neighborhoods for the destination, using your entities and tags as signals plus a location filter The LLM only writes the explanation. Every pick is checked against the results Qloo returned, and anything it invents is dropped. Without Qloo, the app has nothing to recommend.

How we built it

Next.js (App Router), Prisma with PostgreSQL (Neon), Zod validation of the agent's output, Leaflet for the map, an OpenAI-compatible LLM through Groq for the tool-calling loop, and Vercel for hosting. Qloo responses are cached in the database so the demo stays fast and avoids rate limits.

Challenges

Qloo place results include hotels and noisy tags, so we filter and clean them before display. We also handled location objects, timeouts on a free LLM tier, and a fallback that builds a plan from Qloo alone if the LLM is slow or unavailable.

Accomplishments we're proud of

A fully working end-to-end product on a live URL, with an evidence trail on every recommendation and a comparison view that makes Qloo's role obvious.

What we learned

A good recommender needs structured taste signals and location, not just a good language model.

What's next

Group moves for couples and families, more cities, saved taste profiles across moves, and richer neighborhood scoring.

Built With

Share this project:

Updates

Submission history