Lot of work about user requirements notation
Feature language
- gherkin scenarios
Domain model: class diagram that access a domain
Prompt engineering
- zero-shot
- N-shot
- chain-of-thought
Build a domain model with an LLM
LLM
- miss context
- uncertain, unreliable (a second prompt could give you a different answer)
With domain-specific practices:
- the output does not need to be 100% accurate
The same goes with testing: we cannot 100% test a software, it does not even make any sense. The point is: it the output useful?
Workflow representation
- chains
- trees
- graphs