I use promptfoo for our internal evaluations. There are a ton of different "assertions" (tests) that you can write, including model-graded evaluations using rub

I use promptfoo for our internal evaluations. There are a ton of different "assertions" (tests) that you can write, including model-graded evaluations using rubrics. This is far from a solved problem, but there are options out there for systematic testing of LLMs.