RAG

A Retrieval-Augmented Generation (RAG) pipeline in Python: load your own documents, embed them, store the vectors in FAISS, retrieve with hybrid search, and answer questions with a Groq-hosted LLM.

Features

Load PDF, text, and CSV files from a data/ folder (via LangChain loaders)
Chunk documents with RecursiveCharacterTextSplitter
Generate embeddings with sentence-transformers (all-MiniLM-L6-v2)
Store and search vectors with FAISS (cosine / inner product)
Hybrid retrieval: dense FAISS + BM25, fused with RRF, then cross-encoder rerank
Optional LangSmith tracing
Summarize retrieved context with a Groq LLM (llama-3.3-70b-versatile)

Tech stack

LangChain - LangGraph - LangSmith - sentence-transformers - FAISS - BM25 - Groq - pydantic-settings

Project structure

RAG_PRO/
├── app.py                 # Entry point: build store, query, summarize
├── src/
│   ├── config.py          # Settings from .env (models, keys, chunking)
│   ├── data_loader.py     # Load PDF/TXT/CSV into LangChain documents
│   ├── embedding.py       # Chunk documents and create embeddings
│   ├── vector_store.py    # FAISS index: build, save, load, query
│   ├── retriever.py       # Hybrid retrieve + rerank
│   ├── search.py          # RAGSearch: retrieve context + Groq LLM
│   └── graph.py           # LangGraph scaffolding (in progress)
├── data/                  # Your source documents (PDF/TXT/CSV)
├── notebook/              # Exploratory notebooks
├── requirements.txt
└── pyproject.toml

How it works

flowchart LR
    docs[Documents in data/] --> loader[data_loader.py]
    loader --> chunker[embedding.py: chunk + embed]
    chunker --> store[vector_store.py: FAISS index]
    query[User query] --> retrieve[retriever.py: hybrid + rerank]
    store --> retrieve
    retrieve --> llm[search.py: Groq LLM]
    llm --> answer[Answer / summary]

Setup

This project uses uv.

# Install dependencies
uv sync

# Or with pip
pip install -r requirements.txt

Environment variables

Copy the example file and add your Groq API key (get one at https://console.groq.com/keys):

cp .env.example .env

Then edit .env:

GROQ_API_KEY=your_key_here

Optional LangSmith tracing:

LANGSMITH_TRACING=true
LANGSMITH_API_KEY=your_langsmith_key
LANGSMITH_PROJECT=rag-pro

src/config.py requires a real GROQ_API_KEY (not the placeholder) before the app will start.

Usage

Put your documents in the data/ folder (PDF, TXT, or CSV).
Run the pipeline:

uv run python app.py

app.py builds the FAISS store from your documents on the first run, then reuses the saved index on later runs, and prints an LLM-generated summary. Edit the query variable in app.py to ask your own questions.

Notes

The FAISS index is saved to faiss_store/ and is git-ignored (regenerated locally).
Never commit your .env — it holds your API key and is already in .gitignore.

FilesExpand file tree

README.md

Latest commit

History