🧠 Semantic RAG Agent

A context-aware Retrieval-Augmented Generation (RAG) system powered by local LLMs and semantic search.

---

🚀 Overview

This project implements a lightweight yet powerful RAG pipeline that enhances LLM responses using relevant contextual data. It combines embedding-based retrieval with local model inference for efficient and privacy-focused AI.

---

⚙️ Features

• 🔍 Semantic search using SentenceTransformers
• 🧩 Intelligent chunking with overlap strategy
• 📊 Cosine similarity-based retrieval (Top-K matches)
• 🧠 Context injection into LLM prompts
• 💻 Local LLM execution via GPT4All (offline capable)

---

🏗️ Architecture

User Input
   ↓
Chunking
   ↓
Embedding (MiniLM)
   ↓
Cosine Similarity
   ↓
Top-K Context Retrieval
   ↓
Prompt Injection
   ↓
Local LLM (GPT4All)
   ↓
Final Response

---

📁 Project Structure

rag-agent/
│
├── main.py           # Main chat loop + LLM interaction
├── chunks.py         # Chunking + embedding + retrieval logic
├── similarity.py     # Semantic mapping & metadata logic
├── data.json         # Preprocessed knowledge chunks
├── data_set.txt      # Raw dataset

---

⚡ Setup

pip install gpt4all sentence-transformers scikit-learn numpy

---

▶️ Run

python main.py

---

🧠 Models Used

• MiniLM (all-MiniLM-L6-v2) for embeddings
• GPT4All-supported GGUF models (Mistral / LLaMA)

---

📌 Notes

• Large model files (.gguf) are not included in the repository
• Designed for local/offline AI workflows
• Optimized for low-resource environments

---

👨‍💻 Author

Faizan Imran AI Engineer | Full Stack Developer

---

⭐ Future Improvements

• API deployment (FastAPI / Vercel)
• Streaming responses
• Vector database integration (FAISS / Chroma)
• Multi-document support

---