matchmaking agent

Love, simulated. Perfectly matched.

Comment

Inspiration

We were tired of the "swipe fatigue" in modern dating apps. Reducing complex human beings to static photos and bio cards often leads to mismatched expectations and wasted time. We wondered: What if you could test-drive a relationship before saying "hello"? This inspired us to build Matchmaking Agent , a platform where AI digital twins simulate dates on behalf of users to find genuine compatibility based on deep personality traits rather than just surface-level interests.

What it does

Matchmaking Agent creates a "Live Matching Theater" where two users' AI avatars—trained on their Big Five personality traits and core values—engage in simulated conversations.

  • Users can watch these AI interactions unfold in real-time.
  • The Matchmaker AI facilitates the date, asking probing questions and keeping the conversation flowing.

  • The System analyzes the interaction to generate a Compatibility Score ( [ o bj ec tO bj ec t ] S ) and detailed insights, helping users decide if they should connect in real life.

    How we built it

    We architected a real-time simulation engine using FastAPI (Python) for the backend and React for the frontend, connected via WebSockets . The core intelligence is powered by Gemini 3 :

  • Gemini 3 Flash: Powers the User Avatar Agents . Its low latency allows for natural, snappy banter between agents, essential for maintaining the illusion of a real conversation.

  • Gemini 3 Pro: Powers the Matchmaker Agent and Compatibility Analyzer . We rely on its superior reasoning capabilities to understand emotional subtext, detect red flags, and calculate compatibility metrics.

  • AgentScope: We integrated this multi-agent framework to orchestrate the complex state management between the avatars and the environment.

    Challenges we ran into

  • Personality Drift: Early versions of the avatars would revert to being helpful assistants. We had to engineer robust system prompts and utilize Gemini's large context window to maintain consistent "character" traits throughout long simulations.

  • Real-time Synchronization: Managing the state of three autonomous agents (User A, User B, Matchmaker) while broadcasting updates to the frontend via WebSockets required careful handling of async events to prevent race conditions and "double-speaking."

    Accomplishments that we're proud of

  • The "Theater" Experience: We successfully built a UI that feels like watching a movie of your potential relationship, which is both entertaining and insightful.

  • Deep Compatibility Analysis: Moving beyond simple keyword matching to analyzing communication styles and value alignment using Gemini's reasoning capabilities.

  • Seamless Agent Orchestration: Getting three distinct AI personalities to interact naturally without the conversation stalling or becoming repetitive.

    What we learned

    We learned that simulation is the next frontier of social discovery . Static profiles are insufficient for predicting dynamic human chemistry. We also discovered the power of specialized models : using the faster Flash model for conversation and the deeper Pro model for analysis proved to be the perfect architectural pattern for cost-effective yet intelligent agent systems.

What's next for matchmaking agent

  • Voice Integration: Allowing users to hear their avatars speak using real-time voice synthesis.
  • Interactive Interventions: Letting users "whisper" guidance to their avatars during a simulation (e.g., "Ask about their travel habits").
  • Multi-Scenario Testing: Generating diverse scenarios (e.g., "Planning a Trip," "Handling a Conflict") to stress-test compatibility in different life situations.

Built With

Share this project:

Updates