A modular, configurable RAG (Retrieval-Augmented Generation) engine for Data Afrique Hub: upload documents (PDF, TXT, Markdown), ask questions in natural language, and the system retrieves the relevant passages and generates a sourced answer instead of making one up.
Question → Embedding → ChromaDB → Re-ranking → LLM → Answer + sourcesThree deployment modes to fit different constraints (cost, privacy, or performance):
- Free — fully local, no cloud dependency
- Hybrid — local embedding, cloud LLM
- Premium — fully cloud, built for performance
The project also builds on dah-open-rag, the open-source version of the pipeline, released under the MIT license.
Stack
- FastAPI — backend and REST API
- React — interface
- ChromaDB — vector store
- Docker — deploying the different modes