Nice project. Ideally it should be possible to run arbitrary graph or datalog queries to get relevant items but a hierarchical organization w/ basic content type tagging + vector similarity search is a good starting point.

[flagged]