EthnoCAL

Every food tracker I tried seemed to misunderstand the meals I grew up eating. Lentejas con arroz became lentil soup. Ají de gallina became chicken stew.

Before I could log dinner, I had to translate it into something the app understood.

I wanted a food tracker that recognized meals the way people actually eat them, not just as a collection of ingredients. That’s where EthnoCAL started.

The first version of EthnoCAL was a photo-first tracker, much like other nutrition apps. It worked, but I wasn’t convinced it was the right experience.

So I built a chat-first version alongside it. Eventually, I chose the conversation.

Instead of navigating forms, you tell Eth what you ate in English or Spanish. It creates a meal card with a calorie estimate, portion, and any assumptions it made.

I wanted logging to feel more like telling someone about dinner than filling out a form.

One of the biggest decisions was figuring out what AI should actually handle.

EthnoCAL uses a catalog for familiar dishes, with Claude interpreting less predictable conversations. The catalog, not the model, determines nutritional estimates.

Testing also exposed another problem. A phrase like me corté abriendo una lata (I cut myself opening a can) could trigger an inappropriate crisis response.

I tested hundreds of sentences, refined the safety checks, and kept sensitive responses predefined rather than generated by the model.

The goal was to make AI useful without trusting it to make every decision.