Shrewd

Shrewd is a tool that uses large language model (LLM) judgments to create small, efficient local classifiers tailored for fixed text classification tasks. It supports prompt optimization, uses teacher LLMs for labeling, and trains lightweight local student models that run without requiring LLMs at inference time. It is ideal for repeatedly classifying similar text with constrained labeling budgets and inference costs, supporting multiple student model types and decision panel classifiers with calibrated probabilities. It is released as open-source code on GitHub by an independent developer.