An AI-powered assistant for classifying multi-modal documents into Public, Confidential, Highly Sensitive, or Unsafe categories.

• Multi-modal preprocessing for text, PDFs, and images with optional OCR.
• Configurable prompt tree library leveraging LangChain/LlamaIndex orchestration patterns.
• Dual-LLM cross verification with safety keyword detection.
• Citation generation and evidence tracking.
• Human-in-the-loop feedback persistence powered by SQLite and SQLAlchemy.
• FastAPI backend with Streamlit dashboard for interactive and batch workflows.
• Automated report generation in JSON and PDF formats.
• Dockerized deployment, comprehensive tests, and formatting/linting via Makefile.

python -m venv .datathon_env
source .datathon_env/bin/activate
pip install -r requirements.txt

1. Install Ollama on the host where this backend runs and start the daemon with ollama serve (it listens on http://127.0.0.1:11434 by default).
2. Pull the models referenced in config/config.yaml, for example:

   ollama pull llama3.1:8b
   ollama pull llama3.1:13b

3. Optional: verify the service is reachable with curl http://127.0.0.1:11434/api/tags.
4. Adjust config/config.yaml if you need a different model name, base URL, or generation options. The default template configures two local models:

   ollama:
     base_url: http://127.0.0.1:11434
   models:
     primary:
       name: risk-primary-8b
       model: llama3.1:8b
     secondary:
       name: risk-secondary-13b
       model: llama3.1:13b

5. Restart uvicorn (or the Docker container) so the new configuration is loaded. The backend will now call the local Ollama HTTP API for both classifier passes.

1. Authenticate the Ollama CLI (one time):

   ollama signin

2. Register the cloud models you plan to call (examples):

   ollama pull gpt-oss:120b-cloud
   ollama pull deepseek-v3.1:671b-cloud

3. Copy your Ollama Cloud API key from https://ollama.com.
4. Export it before starting the backend (or place it in your shell profile):

   export OLLAMA_API_KEY="sk_live_xxx"

   The default config/config.yaml already points to https://api.ollama.ai and reads the key via api_key_env: OLLAMA_API_KEY, so no further edits are required. If you prefer to hardcode the key instead, replace the env entry with api_key: "sk_live_xxx" (only do this for local testing).
5. Restart uvicorn src.main:app --reload. The classifier will now send requests to Ollama Cloud using your key while retaining the same dual-model workflow.

uvicorn src.main:app --reload

streamlit run ui_dashboard/app.py

make test
make lint
make format

docker build -t doc-assistant .
docker run -p 8000:8000 doc-assistant

See demo.ipynb for an end-to-end walkthrough of the sample test cases (TC1–TC5).