I built this project for the Vancouver AI Data Storytelling Hackathon for rounds 2 and 3. For each free-form question, their responses are plotted as points on a 3D semantic map that are clustered based on meaning. Each cluster is represented by a unique AI voice, and the clusters engage in a roundtable discussion about the question, representing related voices in a concise format.
No followers yet
Once you ship this you can't edit the description of the project, but you'll be able to add more devlogs and re-ship it as you add new features!
Built the project in two parts:
1. Round 2 - semantic maps feature:
- OpenAI text embeddings model for embeddings of survey responses
- t-SNE for collapsing the embeddings to 3 dimensions
- Plotly for plotting the points
2. Round 3 - roundtable discussions feature:
- GPT-5 API for generating discussions with cluster data and vibe instructions
- GPT-4o-mini-tts for generating audio for each dialogue line
- Custom HTML/CSS/JS for audio player and general website design improvements