This project provides a hands-on, educational approach to understanding Retrieval-Augmented Generation (RAG) by building it from scratch using local components. It targets developers seeking a deep, code-level comprehension of RAG pipelines, embeddings, vector search, and context-augmented generation without relying on black-box cloud APIs. The primary benefit is demystifying advanced AI concepts through clear explanations and minimal, well-commented local code.

How It Works

The project meticulously breaks down the RAG pipeline into fundamental steps: data loading, text splitting, embedding generation, vector store creation, retrieval, re-ranking, query preprocessing, prompt augmentation, and final generation. It emphasizes a conceptual clarity and mathematical intuition, starting with simplified examples before introducing vector databases and advanced retrieval strategies. This incremental, code-first methodology ensures users grasp each component's function and interaction within the broader RAG system.

Quick Start & Requirements

• Install: Run npm install in the project root.
• Run: Execute examples using node <path_to_example>/example.js, e.g., node 00_how_rag_works/example.js.
• Prerequisites: Node.js 18+, a local LLM setup (e.g., node-llama-cpp), and necessary npm packages for embeddings and vector math.

Highlighted Details

• Educational Focus: Builds RAG concepts step-by-step with minimal code and clear explanations, avoiding cloud APIs and black boxes.
• Local-First: Designed to run entirely locally, fostering a deeper understanding of underlying mechanisms.
• Modular Design: Features reusable library code for embeddings, vector stores (in-memory, LanceDB, Qdrant), loaders, text splitters, retrievers, and chains.
• Structured Learning Path: Offers a progressive series of examples covering core RAG components from basic similarity search to advanced retrieval strategies and prompt engineering.

Maintenance & Community

Contributions are welcomed for clear, educational RAG examples via pull requests. The project references related concepts and tools like LangChain and AI Agents from Scratch.

Licensing & Compatibility

The provided README does not specify a software license. Potential adopters should verify licensing terms before use, especially concerning commercial applications or integration with closed-source systems.

Limitations & Caveats

This project is primarily educational and under active development, with many advanced features (e.g., hybrid search, multi-modal RAG, production-ready components, evaluation frameworks) still planned. Its focus on foundational understanding means it may not yet offer the robustness or breadth of features found in mature RAG frameworks.