Run API

python -m fastapi dev app/main.py

OCR Backend (Qwen 3 VL 8B Instruct Q4 via Ollama)

This project is configured for Qwen 3 VL 8B Instruct Q4 through Ollama's OpenAI-compatible endpoint.

1. Ensure Ollama is running.
2. Pull the model in Ollama:

ollama pull qwen3-vl:8b-instruct-q4_K_M

3. Set environment variables:
   • OLLAMA_API_BASE=http://127.0.0.1:11434/v1
   • OLLAMA_MODEL=qwen3-vl:8b-instruct-q4_K_M
   • OLLAMA_API_KEY=ollama
   • OLLAMA_TIMEOUT=120
   • OLLAMA_TEMPERATURE=0
4. Run the smoke test:

python examples/basic_usage.py

If Ollama or the model is unavailable, the pipeline falls back to mock mode and logs the exact reason.

Benchmark Raw Folder to SQL Export

This workflow processes each source folder under Benchmark_Data/Raw sequentially, supports both images and PDF files, and writes one SQL file per source folder into Benchmark_Data/SQL.

Generated SQL schema:

• service_ID (globally unique across all generated files)
• service_Name
• service_Origin (source folder name)
• service_Price

Run:

python examples/export_benchmark_sql.py --raw-root ./Benchmark_Data/Raw --sql-dir ./Benchmark_Data/SQL --start-id 1