PDF Chat — RAG Q&A

Ask questions about any PDF using a local RAG pipeline. Upload a document, get instant answers grounded in the text — no hallucinations, no guessing.

Demo

Live demo →

How it works

PDF → text extraction → chunking (500 chars, 100 overlap)
    → sentence-transformers embeddings (all-MiniLM-L6-v2)
    → FAISS index (cosine similarity)

Question → embed → top-4 chunk retrieval
         → LLaMA 3.3 70B (Groq) answers using only retrieved context

1. Parse — pdfplumber extracts raw text from any PDF
2. Chunk — text split into 500-char overlapping chunks
3. Embed — sentence-transformers converts each chunk to a 384-dim vector
4. Index — FAISS stores vectors for fast cosine similarity search
5. Retrieve — at query time, top-4 most relevant chunks are fetched
6. Generate — LLaMA 3.3 70B via Groq answers using only those chunks

Setup

1. Get a free Groq API key

Sign up at console.groq.com — free tier is generous.

2. Clone and install

git clone https://github.com/sa1701/rag-pdf-chat
cd rag-pdf-chat
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

3. Set your API key

cp .env.example .env
# Edit .env and add your GROQ_API_KEY

4. Run

streamlit run app.py

Open http://localhost:8501

Tech stack

Component	Library
UI	Streamlit
PDF parsing	pdfplumber
Embeddings	sentence-transformers (all-MiniLM-L6-v2)
Vector store	FAISS (CPU)
LLM	LLaMA 3.3 70B via Groq API
Orchestration	LangChain

Deploy to Streamlit Cloud (free)

1. Push to GitHub
2. Go to share.streamlit.io
3. Connect repo, set app.py as entrypoint
4. Add GROQ_API_KEY in Secrets settings