Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for Common Ground
Groups never agree on where to eat, and an LLM guessing "something for everyone" gives bland, generic answers. Common Ground is a tool-using agent that treats group taste as a fairness problem and uses Qloo's taste graph to solve it.
How it works
- Each person types a few favourites (artists, films, restaurants, shows, books). The agent resolves each one to a Qloo entity with the Search API.
- For every person it calls Qloo Insights (
urn:entity:place,signal.interests.entities,filter.location.query) to generate dining candidates in the city. - It then asks Qloo to score the same candidate set against each person's taste (
filter.results.entities). The per-person fit bars in the UI are Qloo affinities. - It picks the venue that maximises the worst-off person's fit (maximin), with the mean as tie-break, so nobody is sacrificed for the loudest taste.
- It builds the rest of the evening with cross-domain Qloo Insights (artist, movie, book) using everyone's favourites plus the chosen venue as signals, with explainability turned on.
- An optional LLM only phrases the result; it may not add venues. A visible agent trace lists every tool call.
Why Qloo
The "who likes what, and how much" numbers come from Qloo, not from a language model. Without Qloo the per-person fit scores and cross-domain picks would not exist.
Who it's for
Friend groups, families, teams planning offsites, and event hosts who need a venue that works for mixed tastes.
Notes
Code was written during the submission period. A full live run against hackathon.api.qloo.com passed on 2026-10-20 (3 people, 8 favourites resolved, 3x30 restaurant candidates, 3x50 per-person affinity scores, maximin pick, artist/movie/book extras). The client spaces calls under the 5 req/s hackathon limit and retries 429/5xx. Open source (MIT).
Built With
- javascript
- nebius
- node.js
- openai-api
- qloo-api
- vercel