AI Tutorial
Hands-on notes for the frameworks and ideas people use to build with AI. Each topic starts from zero, then adds real examples, production patterns, and the mistakes that show up once a demo leaves a notebook.
Foundations
Generative AI
What GenAI, ML, and deep learning actually do. Tokens, temperature, hallucinations, embeddings, and a first RAG sketch.
Python for AI
The slice of Python every SDK uses: dicts, functions, JSON, type hints, Pydantic, and async, on one ticket-triage program.
LLM Fundamentals
Tokens, next-token prediction, temperature, prompts, context windows, cost, and latency on a real chat model.
Jev
TypeSafe’s System One model. Choice, score, and yes/no probabilities in one fast pass, so code can route before a chat model is worth calling.
Frameworks and retrieval
LangChain
Models, prompts, tools, and structured output with Pydantic so your code reads fields instead of parsing a sentence.
RAG & RAGAS
Retrieve the right chunks, then score faithfulness and retrieval separately so you know which half to fix.
LlamaIndex
Readers, chunks, indexes, and query engines that answer from your documents and cite the page they used.
Agents, graphs, and tools
LangGraph
Agents as graphs: shared state, loops, branches, memory, and a pause for a human before anything you cannot undo.
Deep Agents
Planning with to-do lists, notes in files, and specialist subagents for jobs that do not fit one prompt.
MCP
The Model Context Protocol: write a tool server once and let Claude, Cursor, and your own agents call it.
OpenAI Agents SDK
An agent loop with tools, handoffs, guardrails, and sessions, built around a shop support assistant.
Claude Code
Anthropic’s terminal agent: read a repo, edit it, run commands, and control it with CLAUDE.md, skills, and hooks.