CommonGround

Inspiration

Making plans with friends should be exciting, but it often turns into a negotiation. Everyone has different interests, favourite places, and ideas of a perfect day. Traditional recommendation systems focus on individuals, while group decisions often end in compromises that nobody is particularly excited about.

We wanted to solve a simple question: What if AI could understand everyone's cultural tastes and discover experiences that genuinely bring people together?

That inspired CommonGround: a social planning experience that uses Qloo's cultural intelligence to turn different preferences into shared adventures.

What it does

CommonGround is a mobile-first, game-inspired social planning application that helps groups discover experiences they can enjoy together.

Users create a group, build individual taste profiles, and select interests across supported cultural categories such as music, movies, books, dining, and places.

Powered by Qloo's Cultural AI API, CommonGround discovers relevant cultural recommendations using individual and group interests.

Its negotiation engine then:

• Combines and deduplicates recommendations from different participants.
• Preserves the cultural evidence behind each suggestion.
• Considers individual preferences and group feedback.
• Respects explicit exclusions and practical constraints.
• Generates an explainable shared experience plan.
• Adapts the plan when someone rejects an activity or changes a requirement.

The result is more than a recommendation list: it is a collaborative experience shaped around the people involved.

How we built it

We built CommonGround using React Native, Expo, and TypeScript for the mobile application, with Python and FastAPI powering the backend.

Qloo's API provides the cultural intelligence layer through entity resolution, cultural insights, and cross-domain recommendations.

The core architecture includes:

• Taste Profiles: Individual participant profiles containing resolved cultural interests.
• Cultural Discovery: Qloo-powered entity search and recommendations.
• Candidate Aggregation: Combines recommendation results while preserving their provenance.
• Negotiation Engine: Uses Reciprocal Rank Fusion, explicit participant feedback, and group-coverage logic to rank candidates.
• Adaptive Planner: Generates, validates, and revises shared plans while respecting hard constraints.
• Local Persistence: Uses AsyncStorage to preserve parties, preferences, negotiation sessions, and plans across application restarts.

We designed the mobile interface around a lighthearted adventure-game aesthetic, with friendly illustrations, pastel colors, rounded cards, and approachable interactions.

We also implemented secure backend-side API authentication, integration health checks, automated testing, and Android and web build configurations.

Challenges we ran into

One of our biggest challenges was combining different recommendation signals without losing the context behind them. When the same candidate appeared in multiple participants' recommendation results, we needed to merge it while preserving where it came from.

Another challenge was distinguishing between ordinary preferences and non-negotiable constraints. Someone being unsure about an activity should not have the same effect as explicitly rejecting it.

We also had to address stale recommendations, asynchronous requests, persistence across application restarts, and partial API failures.

Integrating Qloo required careful handling of entity types, identifiers, authentication, and response structures. We encountered unsupported entity-signal combinations during verification and refined our integration to use valid resolved entities.

These challenges taught us that building a useful agent is not simply about generating recommendations. It is about maintaining trustworthy evidence, respecting user decisions, and adapting reliably.

Accomplishments that we're proud of

• Built a complete mobile-first social planning application.
• Integrated Qloo's Cultural AI API and verified live entity-search and cross-domain recommendation requests.
• Developed individual taste profiles and group-level cultural discovery.
• Implemented a deterministic negotiation engine with recommendation aggregation, participant feedback, and constraint validation.
• Built an adaptive planner that can generate and revise shared experience plans.
• Implemented local persistence and privacy controls.
• Achieved 73 passing backend tests and 175 passing mobile tests.
• Successfully exported Android and web application bundles.
• Established automated CI checks and a containerized backend deployment configuration.

Most importantly, we transformed cultural intelligence into something social: helping people with different interests discover experiences together.

What we learned

We learned that cultural intelligence becomes much more powerful when it is used to connect people rather than simply recommend more things to individuals.

We also learned the importance of distinguishing retrieved evidence from AI-generated interpretations, designing deterministic constraints around agentic workflows, and building systems that handle disagreements instead of hiding them.

Technically, we gained experience integrating a cultural intelligence API, designing recommendation aggregation algorithms, managing asynchronous state, implementing adaptive planning, and preparing a cross-platform application for release.

What's next for CommonGround

We envision CommonGround becoming a companion for shared experiences beyond a single outing.

Future possibilities include real-time group collaboration, shared memories, personalized discovery over time, local event integration, and broader cultural exploration.

We also want to expand the ways groups can discover experiences together while keeping transparency, individual preferences, and genuine cultural connections at the center.

Different people. Different tastes. Common ground.

Built With

• asyncstorage
• docker
• eslint
• expo.io
• fastapi
• github-actions
• google-cloud-run
• jest
• prettier
• pytest
• python
• qloo-api
• react-native
• typescript
• vercel