{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/f77a98e1a02847c595c47ed1c9ac7e16\" frameborder=\"0\" width=\"1440\" height=\"1080\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1080,"width":1440,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1080,"thumbnail_width":1440,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/f77a98e1a02847c595c47ed1c9ac7e16-5fa075a1211fc822.gif","duration":462.119,"title":"LLM Semantic RAG Music Recommender with Guardrails 🎵","description":"This is my CodePath 1.10 final project, an upgraded music recommender that uses a pipeline with vector databases and the Gemini LLM for semantic RAG recommendations and explanations. My main path uses four vector databases plus Gemini, but I added two guardrails in case the API key expires or the profile fails, and I fall back to a semantic-only search, and if the whole pipeline fails I revert to the deterministic recommender from my earlier project. In the demo, I ran three queries like chill acoustic study, aggressive rock, and electronic music and got top recommendations with explanations. I also let users type their own natural query, like somber songs, and it returns a tailored set. No action was specifically requested from viewers."}