Ask your PDF

Retrieval Augmented Generation (RAG) demo, utilizing OgbujiPT and supporting self-hosted LLM.

RAG technique links: overview | paper

UI: Streamlit
- Alternatives: web UI (e.g. Flask), native UI (e.g. Tkinter), cmdline UI (e.g. Blessings)
Vector store: Qdrant
- Alternatives: pgvector, Chroma, Faiss, Weaviate, etc.
PDF to text: PyPDF2
- Alternatives: pdfplumber, pdfreader
- Note: Calibre can be used for e-book cleaning
Text to vector (embedding) model: all-MiniLM-L12-v2.
- Alternatives: See the sentence-transformers pretrained models list, / OpenAI TE ADA 2

Single-PDF support, for now, to keep the demo code simple, though you can easily extend it to e.g. work with multiple docs dropped in a directory.

Note: manual used for demo downloaded from Hasbro via Board Game Capital.

Demo