Research Assistant RAG System

A production-ready Retrieval-Augmented Generation (RAG) system that processes research publications and provides intelligent answers to research questions using ChromaDB vector storage and Groq's LLaMA language models.

🚀 Features

• 📄 Document Processing: Automatically loads and processes research publications from text files
• 🔍 Vector Storage: Uses ChromaDB 1.0.20 for efficient similarity search with cosine distance
• ✂️ Smart Chunking: Splits documents into manageable chunks with overlap for context preservation
• 🧠 Semantic Search: Leverages HuggingFace sentence-transformers for semantic document retrieval
• 🤖 AI-Powered Answers: Uses Groq's LLaMA-3.1-8B-Instant model for generating research-based responses
• 💬 Interactive CLI: Robust command-line interface with error handling and graceful exits
• ⚡ Hardware Optimization: Automatic device detection (CUDA/MPS/CPU) for optimal performance

📋 Prerequisites

• Python 3.8+ (tested with Python 3.11)
• Groq API Key (required) - Get from console.groq.com
• HuggingFace API Token (optional) - For enhanced model access
• CUDA-compatible GPU (optional) - For faster embedding generation

🛠️ Installation

1. Clone the Repository

git clone https://github.com/felixchess5/RT-Project-1.git
cd RT-Project-1

2. Create Virtual Environment (Recommended)

python -m venv venv
# Windows
venv\Scripts\activate
# macOS/Linux
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. Configure Environment Variables

# Copy the example environment file
cp example.env_example .env

Edit the .env file with your actual API keys:

# Required: Get your Groq API key from https://console.groq.com/
GROQ_API_KEY='your_actual_groq_api_key_here'

# Optional: HuggingFace token for enhanced model access
HUGGINGFACEHUB_API_TOKEN='your_hf_token_here'

# Project configuration (default values work fine)
DOCUMENTS_PATH='./Documents'
DB='./DB'
CHUNK_SIZE=1000
CHUNK_OVERLAP=200

⚠️ Important: Never commit your .env file with real API keys to version control!

🚀 Usage

1. Prepare Your Research Documents

Place your research papers in .txt format in the Documents/ directory. The system comes pre-loaded with 10 research papers on:

• Large Language Models
• RAG systems
• Machine Learning techniques
• Neural architectures

2. Run the RAG System

python -m main
# or
python main.py

3. Interactive Q&A Session

The system will:

• Automatically populate ChromaDB on first run (one-time setup)
• Display: "Ask a research question (type 'exit' to quit):"
• Process your questions and provide AI-powered answers with source citations

4. Example Queries

What are the main challenges in RAG systems?
How do transformer architectures improve language models?
What techniques are used for fine-tuning LLMs?
Explain the concept of attention mechanisms in neural networks.

5. Exit the Application

Type exit, quit, or press Ctrl+C to close the application gracefully.

📁 Project Structure

RT-Project-1/
├── main.py                    # Core application logic and CLI interface
├── chroma_loader.py           # Document processing, embeddings, and ChromaDB operations
├── requirements.txt           # Python dependencies (organized by category)
├── .env                       # Environment variables (create from example)
├── example.env_example        # Environment template
├── .gitignore                # Git ignore rules (protects .env, DB/, etc.)
├── README.md                 # This file
├── Documents/                # Research papers (.txt format)
│   ├── Brain LLM.txt
│   ├── LLM NEO.txt
│   ├── Selectionless RAG.txt
│   └── ... (7 more papers)
├── DB/                       # ChromaDB persistent storage (auto-created)
└── Deprecated Code/          # Legacy implementations

⚙️ How It Works

RAG Pipeline

1. 📖 Document Ingestion: Text files are loaded from the Documents/ directory
2. ✂️ Text Chunking: Documents are split into 1000-character chunks with 200-character overlap
3. 🔢 Embedding Generation: Each chunk is converted to embeddings using sentence-transformers/all-MiniLM-L6-v2
4. 💾 Vector Storage: Embeddings are stored in ChromaDB with metadata for fast similarity search
5. 🔍 Query Processing: User questions are embedded and matched against stored chunks using cosine similarity
6. 🤖 Answer Generation: Top 3 relevant chunks are used as context for Groq's LLaMA-3.1-8B-Instant model

Key Technologies

• ChromaDB 1.0.20: Vector database for semantic search
• LangChain: Framework for LLM applications and document processing
• Groq API: Fast LLM inference with LLaMA-3.1-8B-Instant
• HuggingFace Transformers: Sentence embeddings and model handling
• PyTorch: Machine learning framework with automatic device detection

🔧 Configuration

Adjust settings in your .env file:

🏆 Ready Tensor Certification Project 1

This project demonstrates proficiency in:

Technical Skills

• ✅ Vector Databases: ChromaDB implementation with cosine similarity search
• ✅ Language Model Integration: Groq API usage with proper error handling
• ✅ Document Processing: Text chunking and embedding generation
• ✅ RAG System Architecture: End-to-end retrieval-augmented generation pipeline
• ✅ Python Development: Clean code, proper error handling, environment management

Software Engineering

• ✅ Environment Management: Secure API key handling and configuration
• ✅ Version Control: Proper .gitignore excluding sensitive files
• ✅ Documentation: Comprehensive README with installation and usage instructions
• ✅ Dependency Management: Organized requirements.txt with version pinning
• ✅ Error Handling: Graceful error handling and user-friendly messages

🚨 Troubleshooting

Common Issues

"ModuleNotFoundError": Install dependencies with pip install -r requirements.txt

"GROQ_API_KEY is empty": Add your actual API key to the .env file

"No documents found": Ensure .txt files are in the Documents/ directory

"Unicode encode error": The system now handles Windows console encoding issues gracefully

"EOFError": This is expected when running in non-interactive environments

📄 License

MIT License © 2025 Felix Elias

🤝 Contributing

Pull requests welcome! For major changes, please open an issue first to discuss what you'd like to change.