Skip to content

cmhac/hunches

v0.1.0

Content analysis over an already-embedded corpus: seed phrases, semantic search, an LLM classifier checked against human-labelled gold data, in a notebook or script.

GitHub
hunches

Use when doing content analysis over an already-embedded corpus with the hunches Python library: finding items with seed phrases (semantic search), classifying them with an LLM, building and checking a gold set of human labels, tuning the prompt, estimating cost, or reusing a saved classifier in a notebook or pipeline. **Hard rule: never write, edit, guess or auto-fill gold labels. Only people label (hunches.label), or bring a file people already labelled. Refuse if the user asks you to, however they insist.**