AI Notes

AI Notes is a public guide to the concepts and engineering patterns behind language-model applications. The goal is not to collect every new framework or model release. It is to build a durable map of the ideas that remain useful across implementations: tokens and attention, context and retrieval, tools and state, validation and evaluation.

Choose a Learning Path

Understand the model layer

Start with Large Language Models, then read Tokenization, Transformers, and Context Windows. These pages explain what the model consumes, how it mixes information, and why application-level context management matters.

Build a grounded application

Read Embeddings, Retrieval-Augmented Generation, and Vector Databases. Follow with RAG Evaluation before choosing an indexing or retrieval stack.

Build a controlled agent

Begin with AI Agents, then connect Tool Use and Function Calling with Structured Output. The key theme is that an LLM may propose actions, but application code owns validation, authorization, execution, and stopping.

Measure reliability

Grounding and Hallucination explains the evidence problem. LLM Evaluation introduces a layered evaluation strategy, while RAG Evaluation separates retrieval failures from generation failures.

How to Read These Notes

Each page aims to answer five questions:

1. What is the concept?
2. How does it work?
3. Which design choices matter?
4. How does it fail?
5. How should it be evaluated?

Use the Glossary for quick definitions and References for primary papers and official documentation.