A hands-on crash course
A six-session, hands-on crash course that builds from what a language model actually is all the way up to autonomous research agents — wiring API calls, tools, code execution, retrieval, memory, and safety into a working loop. Written for physical scientists who code.
This site renders the six executed notebooks of the LLMs as Agents crash course, with all code and real outputs preserved inline. The material is deliberately provider-general — examples route through OpenRouter so any model can stand in — while LangGraph is featured for building the agent loop.
A single scientific thread — a real spectroscopic analysis — runs through every session, from raw data to a literature-aware research agent. By the end you will have assembled, by hand, the same components that production research agents are made of.
Table of Contents
Written for the SciMLGA 2026 — Scientific Machine Learning for Geophysics & Astronomy summer school.