The job
What the titles mean, what each role is really asked to do, and how to read an advert.
3 of 3 written
AI Engineer, from zero
What the job actually is, and every idea underneath it built from scratch.
Where this stands
31 of 31 lessons written. The bar on every card below is counted from the lesson files that exist, not from a number I typed. A lesson that is planned and unwritten says so on its own page too.
- 130 Python panels, every one executed by Python 3.9.6 during this build
- 812 Python assertions, all passing, or this build would have refused to write
- 1 JavaScript panels you can edit and re-run in the page
- 28 listings shown but not run, each stating why on the panel itself
- No network at read time, and no third-party packages: everything is the standard library
Foundations
The Python, the vectors and the calculus you actually use. Nothing decorative.
4 of 4 written
Classical machine learning
Regression, classification and clustering, each built from scratch and measured.
5 of 5 written
Neural networks
A neuron, a layer, backpropagation, and a network that really trains.
4 of 4 written
How a language model works
Tokens, next-word prediction, attention and a transformer block, from nothing.
5 of 5 written
Building with models
The day job: prompts as contracts, retrieval, agents, and evaluating any of it.
6 of 6 written
Production
Cost, latency, safety, monitoring and the system design questions seniors get asked.
4 of 4 written