hunch

What it is

hunch turns Jev Choice, Score, and Noul questions into Python functions over strings, lists, and pandas DataFrames. It provides verbs like classify, score, check, where, extract, pick, rank, and verify, with request deduplication, caching, and optional LLM escalation for unsure rows.

How it uses Jev

Jev judges each row by answering typed questions: Choice picks one option from a closed set, Score returns a position on ordered levels, and Noul gives the probability a statement is true. hunch sends one request per distinct row, caches answers, and returns plain answers or full distributions with shape labels (sure, split, unsure) that code can branch on.

Primitives:choicescorenoul

Technique worth stealing

One request per distinct row, with several questions about the same row batched together, and answers cached so re-running costs nothing.

Try it

pip install hunch-jev

View on GitHub