Fuse within a unit, chunk across units: joint-vector search for video, photos, slides and any multimodal records on MongoDB Atlas + Voyage AI. Tested on 4 corpora with pre-stated predictions.
Multimodal retrieval engine — index mixed assets (PDFs / Office docs / images), search across four routes (text→text, text→image, image→image, hybrid) fused with RRF. Optional grounded LLM answer layer on top. Local-first FastAPI server with bundled web UI.
