AskMyPDF: RAG-Powered Document Assistant

Overview

Models used

• LLM: Google Gemini 2.5 Flash (via LangChain Google GenAI)
• Embeddings: sentence-transformers/all-mpnet-base-v2 (Hugging Face)

How it works (quick)

1. Load and split data/story.pdf into chunks. 2) Embed and store in a local Chroma DB. 3) Retrieve similar chunks for your question. 4) Ask the LLM to answer using that context.

Setup (uv)

Prereqs: Python 3.13+, uv installed.

1. Create .env with your key:

   GOOGLE_API_KEY=your-google-api-key

2. Put your PDF at data/story.pdf (or change the path in src/main.py).
3. Install deps:

   uv sync

Run

Ask a question from the terminal:

uv run -m src.main "What is SHE doing in India?"

Notes

• Vector DB files are stored in chroma_langchain_db/.
• To force a fresh index, delete that folder and run again.
• Key files: src/main.py, src/vectorstore.py, src/rag.py, src/config.py.
• This is a learning project.