video1.0<iframe src="https://www.loom.com/embed/1f2e344ecf6f4576a834b209e7a88628" frameborder="0" width="1728" height="1296" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12961728Loomhttps://www.loom.com12961728https://cdn.loom.com/sessions/thumbnails/1f2e344ecf6f4576a834b209e7a88628-e0b7a8dbc746c7c4.gif321.645In this Loom I’m recording the main idea of my project, RepoFinder, a document assistant for project planning hosted on GitHub. I start by comparing it to Project 3, the music recommendation system that ranked songs using a largely rule based pipeline and had limits. To address that, RepoFinder attempts to find the right files and returns grounded, actionable resources with citations, using an architecture from user input through an ingestion server and retriever to recommendations and next actions. When I tested it, I saw a lot of negative results, like prioritization and planning responses, and there were issues with retrieval and citation quality. No action was explicitly requested from viewers.