A classifier supporting user-defined criteria. Functions well in content moderation use-cases as well as others such as brand voice checks and prompt injection detection.
It combines classical ML w/ a consensus panel of small open LLMs, which allows for the accuracy associated w/ traditional classification methods, with the generality of LLMs. And because the LLMs have access to Wikipedia and Wolfram, they can navigate evaluations requiring up-to-date factual information as well as math.
