Inspiration
We've all been there: one friend wants a movie, another wants live music, and someone else wants dinner. Recommendation engines usually optimize for one person or average everyone's preferences. But an average recommendation isn't necessarily one that makes everyone happy.
What if an AI agent could negotiate taste instead of just predicting it?
That question inspired TasteTruce, a fairness-first cultural discovery agent powered by Qloo's Taste AI.
Our goal is simple: Different tastes. One great shared experience.
What it does
TasteTruce helps people with different cultural interests discover experiences they can genuinely enjoy together.
Users enter their favorite artists, and TasteTruce:
- Resolves individual tastes: Identifies each artist through Qloo's cultural knowledge graph.
- Explores across domains: Uses Qloo's insights to discover potentially relevant movies, music, and places.
- Finds common ground: Compares recommendations independently generated for each participant and identifies shared entities.
- Negotiates fairly: Ranks shared candidates using a fairness-first scoring algorithm.
- Explains the decision: Presents recommendations with supporting affinity scores, rather than inventing matches.
Unlike traditional recommendation systems, TasteTruce gives greater weight to the person who is least aligned with a proposed experience.
Our consensus score combines 70% of the lowest participant affinity and 30% of the average affinity.
This helps reduce the risk of choosing something that most people like but one person strongly dislikes.
When Qloo doesn't provide enough evidence for a shared recommendation, TasteTruce explicitly reports that no verified match was found.
How we built it
TasteTruce uses Qloo's Taste AI API as its cultural intelligence engine.
The application follows a multi-stage agent workflow:
- Taste resolution: Maps user-provided artist names to Qloo entities using its Search API.
- Individual discovery: Queries Qloo's Insights API separately for each participant.
- Consensus building: Finds candidate entities shared across participant recommendations.
- Fairness optimization: Scores shared results using a weighted minimum-affinity objective.
- Evidence presentation: Displays recommendations with individual affinity evidence and transparent explanations.
The backend is implemented in Python, using its standard library for HTTP communication, API integration, and server execution.
The responsive frontend uses HTML, CSS, and JavaScript.
We use GitHub Actions for automated testing and Render for public hosting.
Qloo API credentials remain on the server and are never exposed in frontend code.
Challenges we ran into
One major challenge was ensuring that recommendations were genuinely relevant rather than simply valid API results.
For example, an early dining recommendation included Brooklyn Bridge, a landmark rather than a restaurant. We responded by adding category-validation logic and regression tests to reject places without explicit dining evidence.
Another challenge was designing recommendations for groups without letting the majority dominate the outcome.
We addressed this through our fairness-first scoring algorithm.
We also had to manage API credentials securely, establish live connectivity, implement request limits, and deploy a working public demonstration.
Accomplishments that we're proud of
- Built a functioning application powered by real Qloo API responses.
- Developed a group-consensus algorithm that emphasizes fairness.
- Implemented cross-domain cultural discovery.
- Added transparent participant-level affinity explanations.
- Built automated tests and a public, open-source repository.
- Deployed TasteTruce to a publicly accessible web application.
- Added safeguards against excessive API consumption.
- Designed the system to acknowledge missing evidence rather than fabricate recommendations.
What we learned
Cultural intelligence becomes more valuable when it helps people make decisions together.
We learned that recommendation quality is not only about relevance or popularity. It also depends on fairness, transparency, and recognizing uncertainty.
Qloo's cultural graph makes it possible to explore relationships across different entertainment and lifestyle domains that would be difficult to infer from individual preferences alone.
What's next for TasteTruce
We want to expand TasteTruce beyond artist-based inputs to support movies, cuisines, books, and richer personal profiles.
Future improvements include stronger location-aware dining recommendations, more sophisticated group negotiation, additional cultural domains, and improved explanations.
Our longer-term vision is to turn TasteTruce into a cultural decision-making assistant for groups of friends, couples, travelers, and event organizers.
TasteTruce isn't just about finding something popular. It's about finding something everyone can enjoy.
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for TasteTruce — Find Common Ground
Built With
- ai-agents
- css3
- github-actions
- html5
- javascript
- python
- qloo-api
- recommendation
- render
- rest-api
Log in or sign up for Devpost to join the conversation.