Prompting as engineering discipline — versioned, measured, and tested — rather than trial and error in a chat window.
What you build
- A prompt library with versioning and a regression test suite
- Structured output that reliably parses, every time
- Chain-of-thought and few-shot patterns benchmarked against each other
- An evaluation harness that scores prompt changes objectively
Format: 10 hours self-paced · Claude, GPT, structured outputs, promptfoo
Proves: you can make model behaviour repeatable enough to ship.
