📚 DocRAG — Local PDF Question Answering with LLaMA-3

Retrieval-Augmented Generation (RAG) over your own PDF documents — powered by FAISS, SentenceTransformers, and Groq's blazing-fast LLaMA-3 inference.

---

✨ What It Does

Drop in any PDF documents, ask questions in plain English, and get grounded, cited answers — without hallucination-prone LLMs making things up. The system retrieves the most relevant passages from your documents before generating a response.

You: "What are the key findings in Q3?"
Bot: "According to page 4 of report.pdf: ..."

---

🏗️ Architecture

                  INGESTION  (ingest.py)                    
  docs/*.pdf  →  PyPDF  →  Text Chunks  →  SentenceTransformers
                                              ↓              
                                     FAISS Index (disk)      


┌─────────────────────────────────────────────────────────────┐
│                   QUERY  (rag_pipeline.py)                  │
│                                                             │
│  User Question  →  Embed Query  →  FAISS Search (top-3)     │
│                                              ↓              │
│  Structured Prompt  →  Groq API (LLaMA-3 8B)  →  Answer     │
└─────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────┐
│                   UI  (app.py)                              │
│                                                             │
│  Streamlit  →  Text Input  →  RAGPipeline.query()           │
│                  ↓                                          │
│     Answer  +  Retrieved Chunks  +  Source Citations        │
└─────────────────────────────────────────────────────────────┘

Component	Technology
PDF Parsing	PyPDF
Embeddings	sentence-transformers (all-MiniLM-L6-v2)
Vector Store	FAISS (persisted to disk)
LLM Inference	Groq API — LLaMA-3 8B
UI	Streamlit

---

🚀 Quickstart

1. Clone & install

git clone https://github.com/your-username/docrag.git
cd docrag
pip install -r requirements.txt

2. Set your Groq API key

export GROQ_API_KEY="your_groq_api_key_here"

Get a free key at console.groq.com.

3. Add your PDFs

mkdir docs
cp /path/to/your/files/*.pdf docs/

4. Ingest documents

python ingest.py

This reads all PDFs in docs/, chunks the text, generates embeddings, and saves a FAISS index to faiss_index/.

5. Launch the app

streamlit run app.py

Open http://localhost:8501 and start asking questions.

---

📁 Project Structure

docrag/
├── docs/                  # 📂 Place your PDF files here
├── faiss_index/           # 🗄️  Auto-generated vector index (git-ignored)
├── ingest.py              # 🔄  PDF → chunks → embeddings → FAISS
├── rag_pipeline.py        # 🧠  Query embedding + retrieval + LLM call
├── app.py                 # 🖥️  Streamlit UI
├── requirements.txt
└── README.md

---

⚙️ Configuration

Key parameters can be tuned at the top of each script:

Parameter	Location	Default	Description
CHUNK_SIZE	ingest.py	512	Characters per text chunk
CHUNK_OVERLAP	ingest.py	64	Overlap between chunks
TOP_K	rag_pipeline.py	3	Number of chunks retrieved per query
MODEL	rag_pipeline.py	llama3-8b-8192	Groq model name
EMBEDDING_MODEL	ingest.py	all-MiniLM-L6-v2	SentenceTransformer model

---

🔍 How It Works (Step by Step)

Ingestion

1. PyPDF reads every .pdf in docs/ and extracts raw text page by page.
2. Text is split into overlapping chunks to preserve context across boundaries.
3. Each chunk is embedded using SentenceTransformers into a 384-dimensional vector.
4. All vectors are stored in a FAISS flat index and saved to disk.

Querying

1. The user's question is embedded with the same model used during ingestion.
2. FAISS performs an approximate nearest-neighbour search, returning the top-3 most semantically similar chunks.
3. Those chunks are injected into a structured prompt alongside the question.
4. Groq's API runs inference on LLaMA-3 8B and streams back the answer.

UI

1. Streamlit renders a text input and a submit button.
2. On submit, RAGPipeline.query() is called and the answer, retrieved chunks, and source filenames are displayed.

---

📦 Requirements

streamlit
pypdf
sentence-transformers
faiss-cpu
groq

Install all at once:

pip install -r requirements.txt

Note: Use faiss-gpu instead of faiss-cpu if you have a CUDA-capable GPU for faster indexing and search.

---

🛠️ Re-indexing

Whenever you add or remove PDFs, re-run ingestion to rebuild the index:

python ingest.py

The old index in faiss_index/ will be overwritten.

---

🤝 Contributing

Contributions are welcome! Please open an issue first to discuss what you'd like to change, then submit a pull request.

1. Fork the repo
2. Create your branch (git checkout -b feature/my-feature)
3. Commit your changes (git commit -m 'Add my feature')
4. Push and open a PR

---

📄 License

This project is licensed under the MIT License. See LICENSE for details.

---

🙏 Acknowledgements

• Groq for ultra-low-latency LLaMA-3 inference
• FAISS by Meta AI Research
• Sentence Transformers by UKP Lab
• Streamlit for the frictionless UI