RAG from Scratch 🔍

Building Retrieval-Augmented Generation (RAG) pipelines from the ground up. Learn how to combine Large Language Models with external knowledge bases for accurate, grounded AI responses.

What is RAG?

RAG enhances LLM responses by retrieving relevant documents from a knowledge base before generating answers, reducing hallucinations and enabling domain-specific Q&A.

Architecture

                    ┌─────────────────┐
                    │   User Query    │
                    └────────┬────────┘
                             │
                    ┌────────▼────────┐
                    │   Embedding     │
                    │   Model         │
                    └────────┬────────┘
                             │
              ┌──────────────▼──────────────┐
              │      Vector Database        │
              │   (Similarity Search)       │
              └──────────────┬──────────────┘
                             │
                    ┌────────▼────────┐
                    │  Top-K Relevant │
                    │  Documents      │
                    └────────┬────────┘
                             │
              ┌──────────────▼──────────────┐
              │     LLM + Context           │
              │  (Generate Answer)          │
              └──────────────┬──────────────┘
                             │
                    ┌────────▼────────┐
                    │  Grounded       │
                    │  Response       │
                    └─────────────────┘

Key Concepts Covered

• Document Loading — Ingesting data from various sources
• Text Splitting — Chunking strategies for optimal retrieval
• Embeddings — Converting text to vector representations
• Vector Stores — Storing and searching document embeddings
• Retrieval Chains — Combining retrieval with LLM generation
• Prompt Engineering — Crafting prompts for grounded responses

Tech Stack

Getting Started

git clone https://github.com/nofaukost/rag-from-scratch.git
cd rag-from-scratch
pip install langchain openai chromadb tiktoken
export OPENAI_API_KEY=your_key_here

License

MIT