📚 RAG Personal Assistant

A personal Retrieval-Augmented Generation (RAG) system that answers user questions over your own documents using either cloud or local LLMs.

This project is designed to demonstrate production-ready patterns for building an end-to-end RAG pipeline, suitable for personal use, local deployment.

---

🚀 Features

✅ Ingest PDF documents into a vector database
✅ Semantic chunking with metadata (e.g., source path)
✅ Local or OpenAI Embeddings (configurable)
✅ FAISS vector store for offline search
✅ Retrieval + Augmented Generation pipeline
✅ Supports:

• OpenAI GPT models (fallback)
• Local GGUF LLMs via llama-cpp-python

✅ Modular and configurable project structure
✅ CLI and notebook demos

---

⚙️ Architecture

[ PDF ] --> [ Chunking & Embeddings ] --> [ FAISS Vector Store ] ↘ [ Retrieval ] ↘ [ LLM ] ↘ [ Answer ]

🗂️ Project Structure

rag-personal-assistant/ ├── models/ # (Put your downloaded GGUF models here) ├── vector_db/ # Local FAISS DB ├── data/ # Your PDFs ├── notebooks/ # Interactive notebooks └── src/ └── rag_assistant/ ├── config.py ├── ingest.py ├── retrieve.py ├── rag_chain.py ├── llm_loader.py └── utils.py

💾 Local Model Support

This project supports local LLM inference using llama-cpp-python.

Recommended model:

• TheBloke/Mistral-7B-Instruct-v0.1-GGUF

✅ Recommended quantization: Q4_K_M (suitable for CPU-only inference)

Download Instructions

1️⃣ Go to Hugging Face link
2️⃣ Download mistral-7b-instruct-v0.1.Q4_K_M.gguf
3️⃣ Place it in your local models/ folder: rag-personal-assistant/ └── models/ └── mistral-7b-instruct-v0.1.Q4_K_M.gguf

✅ Note: This model file is large (several GB) and is NOT included in this repo.

---

🧪 Quickstart

1️⃣ Install dependencies:

pip install -r requirements.txt

2️⃣ Set your .env file:

3️⃣ Ingest your documents:
python -m src.rag_assistant.ingest

4️⃣ Retrieve and answer:
python -m src.rag_assistant.retrieve

5️⃣ Use in notebooks for rapid prototyping!

⚙️ Configuration
All settings are managed in src/rag_assistant/config.py:

* Embedding Model Type (OpenAI vs Local)

* OpenAI model + key

* Local GGUF model path

* FAISS DB path

* Verbose logging

Example:

LLM_TYPE = "local"   # "openai" or "local"
LOCAL_MODEL_PATH = PROJECT_ROOT / "models" / "mistral-7b-instruct-v0.1.Q4_K_M.gguf"

🌐 OpenAI + Local Fallback
✨ This project supports automatic fallback:

Primary: OpenAI GPT (if available)

Fallback: Local GGUF model (if OpenAI is rate-limited / unavailable)

🤖 Why This Project?
This is not "just calling an API" — it demonstrates:

✅ Real retrieval-augmented generation pipeline
✅ Local/offline-friendly architecture
✅ Configurable multi-backend support
✅ Practical engineering for real-world use cases

Perfect for:

Personal assistants

Enterprise knowledge bases

📜 License
Code: MIT License
Content (docs, images): CC-BY-NC 4.0

🙏 Acknowledgements
LangChain

llama-cpp-python

sentence-transformers

FAISS