kevintonb.

2022 · Tonita

Tonita

A client for a natural-language search product built by an ex-Google team on a modern NLP stack.

01 · the problem

A search product built on natural language lives or dies on its client. The whole promise is that an open text box is more capable than the filter grid it replaces, and a shopper will not take that on trust: they will type one sentence, judge the result, and go back to filters if it disappoints. The interface has to make the plain-language path feel like the fast one from the first query, on whatever device the shopper happens to be holding.

02 · what i built

A single cross-platform codebase covering iOS, Android and web, so the search experience is identical wherever a shopper meets it and one team can move it forward. It was delivered against a backend spanning Firebase, Google Cloud and AWS Amplify, which is the kind of split you get when a small team is optimising each piece rather than standardising on one provider.

Questions

What people ask

What does Tonita do?

It is a commerce search product built on natural language understanding: a shopper describes what they are after in ordinary words instead of assembling filters, and the system does the semantic work of matching that against structured and semi-structured catalogue data. The company started in clothing and accessories.

What was built here?

The cross-platform client. One codebase for iOS, Android and web, delivered against a backend spread across Firebase, Google Cloud and AWS Amplify.