A voice app for hands that are always full.

A home daycare provider is one person: caregiver, cook, and record-keeper, often holding a toddler through a twelve-hour day. She also has to answer to three separate sets of rules, all reflected in how the app is built.

None of those rules get followed at a desk, so the paperwork used to happen after the fact, at 8pm, from memory. The whole app exists to move that record-keeping into the moment, hands-free, so the required paperwork falls out of it instead of being redone later.

Before any screen on this page existed, I interviewed three home-based providers separately: Angie, Deborah, and Yari. I also talked with the administrators who process food-program reimbursements, and with childcare influencers who talk to that audience online.

All three providers, without hearing each other, described the same problem: watching children all day leaves no attention left for the paperwork it also requires. And they described the same fix: capture it by voice in the moment, then review it later, instead of trying to remember it at night. That's what the Voiceboard and the review queue described below are built around.

The main screen is the Voiceboard: one row per child, one column per event type, for a single day. A provider scanning twelve children needs the rows to scroll left and right together, so they stay lined up on the same point in time. Without that, the grid is twelve unrelated timelines that happen to be stacked.

When the app can't understand what was said, it saves the attempt instead of throwing it away, along with the raw words, a translation, and the recording. A badge shows how many are waiting for review.

On the review screen, the words appear as clickable chips. Tapping the one that was the child's name teaches the app that name, and the same mishearing gets recognized correctly every time after. A single correction keeps paying off well beyond the one record it fixed.

The app also shows what it partly understood instead of discarding it, so a provider only fills in the missing piece. And instead of announcing a failure out loud in a room full of children, the spoken reply just says “processing unknown event for unrecognized child”, while it waits for review.

A wake-word detector runs directly on the device, always listening for "log a tot", so no audio ever leaves the device just to hear that phrase. With a live microphone in a room full of other people's children, that was the first question every provider asked.

Turning speech into a saved record runs on a fixed set of rules, with no AI guessing in that step. What's said is checked against a list of known phrases, and names are matched against the class roster. An AI model making that call live would be slower and could confidently guess wrong, logging a nap that never happened. The rule-based system either finds a match or sends it to a person to review, and a non-match is treated as normal, not a bug.

AI is used earlier instead. When a child is added, it generates a list of ways that name is likely to be mispronounced or misheard, and the app checks new recordings against that list. Toddler names are short and similar to each other, so getting them right is the app's hardest problem. Corrections a provider makes in the review queue get added to that same list, so the system keeps improving.

Spanish speech gets translated to English before the app tries to match it, so one set of phrases covers both languages. A single child mentioned updates the screen right away; several children at once take a moment longer in the background, so the response never slows down.

The paperwork stopped being a second job. Attendance records, signed releases, food-program claims, and safety logs all come from events already captured during the day, so month end became a quick review instead of an evening of catching up. The app shipped as a paid product, sold directly and through childcare associations, but the number I'd point to first is the evening a provider got back.