Jewish Studies · AI

Source Sheet Creator

Type a topic or question. Candidate sources come from Sefaria's topic pages and full-text search, TypeSafe's Jev model judges each one, and the best are laid out as a printable Hebrew–English sheet.

Try:

How it works

Gather. The key terms of your request are looked up as Sefaria topic names and searched as full text, both together and in adjacent pairs.
Pick topics. Name matching is fuzzy: "abusive" can surface Rabbi Akiva. Jev answers one yes/no question per matched topic ("is this a main subject of the request?") in a single call, and only those topics' curated sources are kept.
Judge. Up to 60 candidates are fetched in Hebrew and English. For each one, Jev rates relevance (unrelated → passing mention → related → central) and whether the passage is substantive rather than a heading or list of citations.
Select in code. Rank = relevance/3 × (0.4 + 0.6 × substantive). The sheet keeps anything rated at least "related", allows at most two sources per book, drops duplicate passages, and groups sources by Tanakh, Mishnah, Talmud, Midrash, Halakhah and so on.

Jev returns typed judgments and calibrated probabilities rather than generated text. The model only judges sources; it never writes or paraphrases them. Every word on the sheet comes from Sefaria. The same pipeline also runs locally as a Python script behind a one-click Mac Dock app.