A scanned identifier becomes one clean, reviewed record instead of four contradictory search results. The identifier is checksum-gated, fanned out in parallel to several external data sources, and the competing candidates are merged by a language model into a single draft a person approves. When no identifier is readable, vision on the captured photo takes over, and results without an identifier are cached by normalised name.
What ports as-is
- The fan-out and merge engine
- The checksum gate
- The vision fallback path
- The name-normalised cache
- The draft-review contract
- Server-side barcode detection on uploaded photos with preprocessing variants
What we build for you
- The external data sources
- Which are the whole per-customer question: the wired-in set fits consumer goods
- So parts
- Assets
- Documents or any other domain needs its own sources and its own merge rules
Plus the fields your record must carry.
