MODEL 03 · NAMED-ENTITY TYPING
Decide whether a string is a personal name, a company, a place, a brand or product, or none of the above. The model returns calibrated probabilities — so you can route the long tail of CRM and KYC noise to the right pipeline before downstream parsing kicks in. The hybrid CNN-Transformer follow-up is published as Onomas-CNN X: 92.1% accuracy at 2,813 names/sec on a single CPU.
Paste a string from a form field, a CRM record, or anywhere else. The classifier returns probabilities across the five-way label and the chosen top class.
RELATED
Once you know it's a person name, parse it into given names, surnames, particles, titles.
For person names: predict gender from the given-name fragment.
If the input is a person name in a non-Latin script, render it for downstream consumers.