Semantic forensic search that bridges family descriptions and unidentified-remains records.
- Python
- SapBERT
- Actian VectorAI DB
- Named Vectors
- NamUs
- Next.js
THE PROBLEM
Missing-persons cases stall because the way a family describes someone almost never matches the clinical language in a forensic record. A mom says 'eagle tattoo on his left arm' and the case file says 'avian motif dermagraphic.' Keyword search misses that link every time, so real matches sit buried in the federal NamUs database.
THE BUILD
I built a search engine that bridges the two vocabularies. It runs SapBERT medical embeddings and named vectors over 40,000+ NamUs records, then fuses structured filters with vector similarity so a plain-English description surfaces the right forensic record. The whole thing runs on Actian's VectorAI database.
THE RESULT
Trace took 1st place out of 22 teams at the Actian VectorAI Build Challenge, announced at AI Dev '26 in San Francisco. It is live and public, and a family member can search it in the language they actually use.