Documentation

Pages in this documentation

• Quick Start, Evaluation, & Development (current page)
• Architecture & Self-Hosting
• Services Reference
• API Reference

Multilingual text analysis platform for detecting bias indicators in written content. Combines rule-based lexicon matching, ML models, and LLM integration.

Supported languages: English, Portuguese, Czech, Finnish, Greek

Quick Start

1. Copy .secrets.sample to .secrets and set your API keys:

   • LLM_API_KEY.txt - API key for local self-hosted LLM
   • GEMINI_API_KEY.txt - API key for Google Gemini (optional)

2. Build and run all services:

   docker compose up -d --build

3. Open http://localhost:7020 in your browser.

Selective Startup

To save resources or focus on specific features, you can start only the services you need. For example, to run just the web interface and rule-based detection:

docker compose up -d web gateway rule-based-service

Development

For development with hot reload and debugging:

docker compose -f compose.yml -f compose.dev.yml up -d

This enables hot reload for the frontend and debugpy on port 5678 for backend services.

Running Evaluations

The evaluation/ folder contains scripts for measuring service accuracy. These run outside Docker using UV:

cd evaluation
uv sync
uv run python evaluate_rule_based.py
uv run python evaluate_llm.py

See evaluation/README.md for detailed instructions.