GPU Data Science Tutor - Self-Corrective RAG System

A comprehensive Retrieval-Augmented Generation (RAG) system specifically designed for GPU-accelerated data science education. The system combines web documentation, Jupyter notebooks, and Python code examples to provide accurate, well-cited responses about CUDA programming, RAPIDS, PyTorch, TensorFlow, and other GPU computing frameworks.

🤝 Contributors

⚙️ Soham Sarkar1, ⚙️ Omik Save2, 📁 Arjun Shilamkoti3

1 Developer, School of Electrical, Computer and Energy Engineering, Arizona State University
2 Developer, School for Engineering of Matter, Transport, and Energy, Arizona State University
3 Content Manager, W. P. Carey School of Business, Arizona State University

🚀 Features

🔍 Multi-Source Knowledge Base

• Loads 90+ GPU data science URLs from CSV
• Integrates Jupyter notebooks (.ipynb) and Python files (.py)
• Supports mixed citation formats for different source types

🧠 Self-Corrective RAG Pipeline

• Enhanced Retrieval: Multi-attempt document retrieval with relevance grading
• Hallucination Detection: Validates responses against source material
• Answer Quality Assessment: Ensures responses adequately address questions
• Automatic Regeneration: Improves responses through iterative refinement
• Verified Sources: Cites the sources used in content creation from the knowledge base

📚 Intelligent Citation System

• Numbered citations [1], [2], [3] with proper source attribution
• Reference validation to prevent hallucinated sources
• Automatic reference section generation

💬 Conversation Management

• Chat history tracking for follow-up questions
• Context-aware response generation
• Document exclusion to prevent repetitive retrievals

🛠️ System Architecture

📁 Project Structure

File / Folder	Description
README.md	You're here! This file describes the project, components, and how to get started.
rag_self_corrective.ipynb	🔧 Main Jupyter Notebook – Core implementation of the RAG-based chat interface with self-reflection and answer validation.
rag_compare_models_naive_rag_vs_base_model.ipynb	🔧 First implementation comparing LLama3 8B base model and RAG-based model.
requirements.txt	📦 Python dependencies required to run the notebook and supporting code.
gpu_data_science_urls.txt	🌐 List of curated URLs used as the knowledge base for retrieval (GPU and data science related).
resources/	📚 Supplementary files, documents, and articles used for building or augmenting the knowledge base.

---