video1.0<iframe src="https://www.loom.com/embed/cf1d8eae22e943d883aae4ed3f05db10" frameborder="0" width="1600" height="1200" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12001600Loomhttps://www.loom.com12001600https://cdn.loom.com/sessions/thumbnails/cf1d8eae22e943d883aae4ed3f05db10-e247fcf934cb9eb1.gif300.988I built a song recommendation system using a rack style flow and a RAG engine. My project connects to a Supabase database, and the main logic is in a Python recommended file. I use mood and genre as features, and the dataset is 18 songs stored with an embedding vector database built from a CSV. The output is generated by Gemini, and I included a system diagram plus unit and integration tests with pytest, including edge cases like missing context. I asked you to run the pytest command if you want to verify it, though it may be slow at first.