Just talk: voice food logging is here

This one is the feature I built the rest of the app to support.

Open the food log page, hit the microphone button, and just say what you ate. “Two scrambled eggs, a slice of sourdough toast with butter, and a flat white.” The mic listens in real time, transcribes as you go, and as soon as you stop talking it hands the transcript to the AI. A few seconds later you get a nutrition card back with the items, quantities, and macros laid out cleanly.

You can accept it as is. You can edit any quantity if it got something a bit off. Or you can reject the whole thing and try again. There is no commit until you press accept, so nothing weird ends up in your log because the AI misheard you.

Under the hood this is the OpenAI Realtime API. I went back and forth on whether to wait for the API to mature, but the latency is good enough now that the experience feels like the right one. Tap, talk, see a card, decide.

A couple of things I learned shipping this. Most voice apps make you wait through a noticeable beat before anything shows up. That beat is where people start second-guessing whether the app heard them. Streaming the transcript while you are still talking turned out to be the single biggest comfort improvement. You can see the words land.

The other piece is the language thing. Realtime APIs love to drift into other languages if you give them silence or weird audio. I lock the conversation to English for now so that does not happen by accident.

If voice is your thing, the rest of the app got reorganized around it. The mic button is the centerpiece of the food log page, with the recent history sitting just below it.

by Boyd Thomson