{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/1c2b2d221552456994a9b9e7cd46fe7c\" frameborder=\"0\" width=\"1728\" height=\"1296\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1296,"width":1728,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1296,"thumbnail_width":1728,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/1c2b2d221552456994a9b9e7cd46fe7c-17969ebfc3ddb6d3.gif","duration":696.884,"title":"How MelodyMind Explains Music Recommendations","description":"I built MelodyMind, an AI music recommendation agent that reasons and explains, fixing my earlier rule based module 3 system that only matched labels. The agent runs six steps, starting with a music guardrail, then extracting your intent, doing vector search in ChromaDB for top 10 songs, pulling genre mode context from a knowledge base, having the LLM pick best 5 with explanations and tradeoffs, and saving a full log trace. In a test, it recommended late night lofi with clear reasons and handled conflicting high energy and sad mood by lowering confidence and explaining the trade off. It also refused non music input. I asked viewers to click Recommend during the live demo, and I ran an evaluation showing improved diversity in 4 out of 5 cases versus the baseline."}