Why this is in the programme
The final test of an AI system is not whether it produces impressive outputs, but whether it satisfies a useful requirement sufficiently well to justify its use.
Evaluation therefore needs to begin with the problem rather than the model.
Course
Develop and Evaluate LLM Features Effectively — Coursera
The course covers requirements, product definition, acceptance testing, user acceptance and the use of measurable criteria for evaluating LLM features.
Role in the experiment
This stage will help define the evidence needed to judge the eventual system.
The objective is to finish the twelve-week experiment with more than a subjective judgement that the prototype "works". Its value and limitations should be supported by explicit acceptance criteria and measured results.