<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/f02d818fa4ff4a43bfe97ec3686fe4ae&quot; frameborder=&quot;0&quot; width=&quot;1728&quot; height=&quot;1296&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>1296</height><width>1728</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>1296</thumbnail_height><thumbnail_width>1728</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/f02d818fa4ff4a43bfe97ec3686fe4ae-9036515f078fbae3.gif</thumbnail_url><duration>169.25866699999997</duration><title>Deterministic AI for Legal Research and Drafting</title><description>This Loom presents Bucket, an AI workspace designed for India’s 1.5 million advocates working in under-resourced districts and lower courts. It addresses the risk of generic legal AI fabricating authorities, referencing a February AI petition by India’s Chief Justice that used a completely fabricated case as binding authority. Bucket claims to auto-generate case arguments and predict likely bench questions with deterministic, traceable outputs grounded in a 20 million judgment library. Its architecture uses a five-layer pipeline including deterministic scoring, retrieval from top five precedents via Pinecone, iterative relevance grading, and a final generation step that traces every claim to specific sourced clauses.</description></oembed>