Large language models
Eleven weeks on the gap between a demo and a system: prompting as a contract, retrieval that earns its place, fine-tuning only when it pays, and a cost budget you can defend.
- Length
- 11 weeks
- Time
- 4 h a week
- Enrolled
- 2,148
- Next cohort
- January 2027 cohort
- Written and marked by
- Jonas Reiter Lead instructor, Systems
Four modules, one assignment each
Every module ends with work that is read by the person who wrote it. The assignments build on each other, so the order is fixed.
- 01
Prompting as an interface contract
A prompt is a specification other people will maintain. Writing one that survives a model upgrade, and testing it like code.
- 02
Retrieval and grounding
When a long context window is not memory. Chunking, ranking, and the evaluation that tells you whether retrieval helped at all.
- 03
Fine-tuning: when it pays
Three cases where it does, seven where a better prompt or a smaller model does the same job for a tenth of the cost.
- 04
Cost and latency budgets
Tokens are money and seconds are churn. Setting a budget per request before the architecture is chosen, and holding to it.
Who it is for
Engineers and technical product people who have already built something with a model and found the second version harder than the first. You have a working prototype and a list of things it gets wrong that you cannot quite explain.
It is not an introduction to machine learning, and it does not train models from scratch. If you want to understand transformers from the inside, this is the wrong track and the scoping call will say so.
How the eleven weeks run
Two taught hours a week, live and recorded. Two hours of your own on the assignment, which changes every fortnight and is read by the instructor who wrote the module. The assignments build on each other: the prompt contract from week two is what you evaluate in week six and what you put a budget on in week nine.
The final brief comes from a partner company, or from your own work if you have a system that needs this. You defend it in front of two practitioners, one of whom will ask what it costs to run.
