Opinion Visualization with Annotated Labels — an experimental tool for visualizing public opinion data.

Users upload CSV files of comments and vote matrices (or a single h5ad file), annotate representative comments on a scale, and the system uses PCA + linear regression to score and visualize all comments along user-defined dimensions.

_{Data shown from pol.is/report/r7cwmiaxczyj8te9rzdmx — exported via valency-anndata Streamlit app}

• Node.js (v18+)
• uv (Python package manager)

make install   # Install frontend and backend dependencies
make dev       # Start both dev servers

The frontend runs at http://localhost:5173 and the backend at http://localhost:8000.

make install   Install all dependencies
make dev       Start both frontend and backend dev servers
make build     Build frontend for production
make lint      Lint frontend code

Upload two files:

• Comments CSV — must contain comment-id and comment-body columns
• Votes matrix CSV — must contain a participant column, with remaining columns as comment IDs and values as votes

Upload a single .h5ad file following the valency-anndata data model:

• .X — votes matrix (participants x statements), values {-1, +1, NaN}
• .var['content'] — statement/comment text

The h5ad file is parsed client-side and converted to the CSV format automatically. You can generate h5ad files from Pol.is report URLs using the valency-anndata export app.

• demo/oval/ — React/TypeScript frontend (Vite + Tailwind CSS)
• server/ — FastAPI backend (PCA, regression, scoring)
• data/ — Runtime data directory (gitignored)

See CLAUDE.md for detailed architecture documentation.