video1.0<iframe src="https://www.loom.com/embed/5c23f0f2085a4657a1b645b6720eea58" frameborder="0" width="1662" height="1246" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12461662Loomhttps://www.loom.com12461662https://cdn.loom.com/sessions/thumbnails/5c23f0f2085a4657a1b645b6720eea58-d2e341c1278351d4.gif492.825In this CodePath Applied AI System project, I expand a deterministic music recommender into an agentic system that can reason about its own output. I used a plan act check and repair loop, with guardrails that score confidence and flag issues like conflicting signals or low diversity, then retry with an adjusted strategy while keeping explanations transparent. In agent mode I demo genre first, chill lo fi, and intense rock with messy inputs. Evaluation across test scenarios passes with an average confidence of 0.72, and unit tests pass. No viewer action was requested.