Ontology 1 wants to understand what buyers are really looking for

Onton's Ontology 1 model interprets complex and contradictory queries without being limited to the keywords of home decor product sheets.

Named Ontology 1, Onton’s new model tackles searches that are difficult to formulate using traditional categories or filters. Queries like "a rug that hides cat vomit but isn't beige" or "a laundry basket I won't hate looking at for ten years" are interpreted based on intent, desired style, and implicit constraints.

Its neurosymbolic architecture combines statistical learning with reasoning from a structured knowledge base. Instead of limiting itself to the words present in product descriptions, the system links their properties to verify if the result truly matches the request. To search for a pet-friendly sofa, for example, it can examine the fabric, its durability, or how easy it is to clean, even if the term "pet-friendly" does not appear in the description.

Onton also claims that the model progressively enriches its knowledge from queries without requiring a full retraining. The relationships learned for one type of product can then be reused for other searches. The system can process text, images, mood boards, or incomplete and sometimes contradictory information.

In an internal test consisting of 90 complex searches in the home decor space, Ontology 1 achieved an average precision of 63% among the top ten results, compared to 54.3% for Google Shopping and 46.9% for Amazon. It won 52 queries, despite a catalog reported to be significantly smaller.