{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/348e3c72c98a46aebd4932a529b1ca9a\" frameborder=\"0\" width=\"2068\" height=\"1551\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1551,"width":2068,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1551,"thumbnail_width":2068,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/348e3c72c98a46aebd4932a529b1ca9a-9f3b221d0f84b8d2.gif","duration":508.373,"title":"Automatically Scoring Docs Before Human Review","description":"This Loom demonstrates how to use LLM criteria and scoring to produce shipping-ready onboarding documents. Lucy Bursey explains that her system uses a rubric-based approach with two writing stages, a third stage that scores outputs against an editable, customizable rubric, and a fourth stage that revises failures, stopping after three attempts per stage with detailed logs. She emphasizes that the rubric is calibrated to real PRD requirements, not generic “good vibes,” and highlights a prior failure where the model missed a risk coverage section because the standard PRD lacked one. She notes there is a threshold of six criteria and that automation should catch issues before humans review, including integration failures that happen at off hours."}