🧩 Overview
This project implements a Retrieval-Augmented Generation (RAG) pipeline that allows a local Large Language Model (LLM) — such as mistral:7b-instruct — to answer user questions based only on embedded local documents, not hallucinated data.
It is designed as a fully offline AI assistant, using:
Ollama for local LLM inference
FAISS for vector search
NumPy for embedding vector storage and similarity retrieval
The system retrieves the top relevant text segments from your dataset, feeds them into the LLM context, and generates a grounded answer.
⚙️ Features
✅ Local & offline — no OpenAI API key required
✅ FAISS vector search for document retrieval
✅ Ollama integration with any local model (mistral, llama3, gemma3, etc.)
✅ Context truncation handling (prevents model overload)
✅ Grounding verification — detects if answer was based on retrieved context
📁 Project Structure
rag-assistant/
├── data/sample_docs/ # Your local corpus (Wikipedia or Ready Tensor publications)
│ ├── RAG.txt
│ ├── LongChain.txt
│ ├── MCP.txt
│ └── Agentic AI.txt
├── src/
│ ├── embeddings.py # Local embedding generator
│ ├── indexer.py # FAISS index builder / loader
│ ├── generator.py # Model interaction (Ollama)
│ └── pipeline.py # Main RAG pipeline (retrieval → generation → grounding)
├── index/ # Auto-generated FAISS vector index (ignored by .gitignore)
├── .env_example # Environment setup template ✅
├── .gitignore # Secure Git ignore configuration ✅
├── requirements.txt # Python dependencies
└── README.md # This documentation
🚀 Setup Instructions
1️⃣ Install Requirements
Make sure Python 3.10+ is installed, then run:
pip install -r requirements.txt
2️⃣ Install Ollama & Model
Download and install Ollama:
👉 https://ollama.com/download
Then pull a compatible model (recommended):
ollama pull mistral:7b-instruct
3️⃣ Prepare Environment
Copy the example environment file:
cp .env_example .env
You may adjust:
OLLAMA_MODEL=mistral:7b-instruct
🧠 How to Run
Build Index
python -m src.indexer
Ask a Question
python -m src.pipeline --ask "What is RAG?"
✅ Expected Output:
🔍 Retrieved context sources:
💬 Model Answer:
Retrieval-Augmented Generation (RAG) is a technique...
✅ Answer appears grounded in retrieved context.
🔒 Environment & Security Practices
This repository strictly follows Ready Tensor Secure AI Development guidelines:
File Purpose
.gitignore Prevents sensitive or large files (e.g., .env, index/) from being uploaded
.env_example Documents required environment variables without exposing real data
.env Private local file containing your runtime configuration — never committed
📜 Documentation & Reproducibility
Clear file structure and reproducible setup
No proprietary dependencies (fully open-source)
Runs 100% locally with Ollama and FAISS
Meets Ready Tensor’s Technical Rubric for “Functional RAG system” and “Best Practices for AI/ML Documentation”
🧾 Licensing & Data Source
This project uses Wikipedia articles for demonstration.
All content complies with Wikipedia’s CC BY-SA 4.0 License
.
If you adapt or expand this system using Ready Tensor publications, ensure authors permit reuse under Ready Tensor’s platform terms.