An employee scans a product, and four conflicting search results become one verified record. The search runs in parallel across several data sources, a language model merges the matches into a single proposal, and a person confirms it before it enters the system.
When the code on the label is damaged or missing altogether, image recognition takes over: a photo of the product is enough to identify it anyway. Results with no readable code are remembered by name, so the same unsuccessful search never runs twice.
The result: your staff member no longer compares four conflicting results from four sources by hand, but confirms one finished proposal instead.
What ports as-is
- Matching several data sources into a single proposal
- Checking the scanned code is valid before the search starts
- Image recognition that steps in when no code is readable
- The cache that recognises an already-seen product by name
- Nothing is taken over until a person confirms it
- Detecting barcodes from uploaded photos, with several preprocessing steps running in the background
- A second product database that steps in automatically when the first finds nothing
- A notice about a new version that invites a reload, instead of quietly leaving the app running on an outdated version
What we build for you
- The external data sources themselves
- The built-in set fits consumer goods, which is the real question per customer
- Spare parts need their own sources
- Assets need their own sources too
- Documents or any other domain need their own sources and their own rules for merging matches
Plus the fields your record must carry.
