shubhamsaboo/all-rag-techniques

All-rag-techniques is a learning resource that breaks down how AI systems can answer questions using your own documents, instead of relying solely on what the AI was trained on. This concept is called Retrieval-Augmented Generation, or RAG. The repo provides a collection of individual Jupyter Notebooks, each explaining a specific technique for making this process work better. Instead of using heavy frameworks that hide the details, it uses plain Python so you can actually read, understand, and modify the code. RAG works by taking a user's question, searching a set of documents for the most relevant text, and then passing that text to a language model to generate an answer. The challenge is that basic search often misses context or retrieves the wrong information. This project demonstrates over 20 different strategies to fix that. For example, some notebooks show how to split documents into better chunks, how to rewrite a user's question for better search results, or how to combine text and images for richer answers. Each notebook includes explanations, code, visualizations, and evaluations so you can see how each technique performs. This is ideal for developers, product managers, or founders who want to understand how RAG actually works under the hood. If you are building an AI chatbot for customer support, a tool that searches through company PDFs, or any application that needs an AI to reason over specific data, these notebooks show you the mechanics step-by-step. It is also useful if you already use tools that do RAG for you, but you want to understand the tradeoffs and options available so you can make better decisions. A notable tradeoff is that the project intentionally avoids popular frameworks. This makes the code more readable and educational, but means you would need to build out more robust infrastructure if you wanted to move these techniques into a production system.