Using AI to identify narrative passages in the Babylonian Talmud
The Babylonian Talmud is filled with stories—narratives about rabbis, their lives, their teachings, and the world around them. These stories are woven throughout the Talmud's legal discussions and are treasures of Jewish literature.
But there's a problem: these stories are scattered throughout the Talmud with no index. They appear without warning between legal debates, sometimes just a few sentences, sometimes spanning pages. Scholars spend years learning to recognize them.
Our Goal: Create the first comprehensive catalog of narrative stories in the Talmud, making them accessible to scholars, students, and curious readers everywhere.
Finding stories in the Talmud isn't a simple search problem. Here's why:
Stories don't announce themselves. The same Hebrew words appear in both legal and narrative contexts.
Stories range from two sentences to multiple pages, with no standard format.
Narratives are interwoven with legal discussions, often mid-paragraph.
We're starting with Tractate Ketubot and expanding from there.
Each round of expert feedback transformed the system. What started with a 50% false positive rate became a 96.3% accurate pipeline—through eight versions of learning what makes a story.
The biggest breakthrough came from expert validation. When Talmud scholar Jeffrey Rubenstein reviewed our first results, he found the AI was confusing legal attribution ("Rabbi X quotes Rabbi Y") with characters in a story. That single insight eliminated 53 false positives. Every version since has been shaped by his corrections.