{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/54683866a75b4fe5bf4342d85c958a45\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/54683866a75b4fe5bf4342d85c958a45-f069ceeab6c8a164.gif","duration":345.09,"title":"Round 2","description":"This Loom walks through Shaurya’s approach to tracking persona drift in a fully offline-first adaptive rack system. The raw dataset lacked real-time timestamps, so Shaurya implemented a synthetic chronological clock that bucketed every 15 messages into a simulated day to analyze sentiment and tone across 734 continuous days. To detect drift triggers without heavy LLM summarization, he used a lightweight TF-IDF keyword extractor combined with relationship tags to isolate what caused tone shifts."}