VibeCheck

Inspiration

The motivation came from the exhaustion of modern dating. Swiping through endless profiles and suffering through repetitive small talk feels like a chore. We wanted to see if we could automate the "vibe check" phase to help people skip the boring parts and get straight to the chemistry.

What it does

VibeCheck uses AI to simulate how two people would actually interact. Users answer a series of trivial questions that reveal their personality subtext. When a match happens, two AI agents—digital twins—engage in a three-round banter session. The app then provides a transcript of that chat along with a "Vibe Audit" that scores the compatibility and predicts if the match would actually work in real life.

How we built it

The stack is built on a React and Vite frontend with a FastAPI backend. We used the Strands SDK to orchestrate the multi-agent conversation via Amazon Bedrock. Amazon Nova Lite handles the fast-paced chat turns while Amazon Nova Pro performs the deep personality synthesis and the final audit. Data is stored in a SQLite database, and the entire system is hosted as a monorepo on Vercel.

Challenges we ran into

Deployment was a major hurdle. We faced persistent CMake compilation errors while trying to install mathematical libraries in a serverless environment. We had to carefully manage Python runtimes and namespace collisions to ensure our local agent logic wasn't being overridden by external packages. Additionally, we spent significant time engineering prompts to prevent agents from breaking character during the automated chat.

Accomplishments that we're proud of

We successfully built a system where two distinct AI personalities can have a coherent, sarcastic, and revealing conversation based on nothing but a few MCQ answers. Watching the "Vibe Supervisor" accurately roast the interaction was a high point for the team.

What we learned

We learned that the intangible "vibe" of a conversation can actually be quantified. By analyzing markers like spontaneity, curiosity, and friction, we found that AI can effectively predict social resonance. We also gained deep experience in managing complex Python dependencies in serverless architectures.

What's next for VibeCheck

We plan to add voice synthesis so users can hear their agents talk. We also want to implement a feature where users can "whisper" to their agent mid-conversation to nudge the direction of the chat.

Built With

• fastapi
• javascript
• python
• react
• strandssdk
• vite