PeetaBoo was built by a parent, for parents, because the most important moments happen when you're not watching. It's named after the way our daughter Ellie says peekaboo.
It started with a wall-mounted security camera in our living room and a hard drive filling up with one-minute motion clips. Thousands of them. Ellie's first steps were in there somewhere. So were her first words, her first climb onto the armchair, and a lot of peekaboo. Almost none of it had been watched, because nobody has time to scrub through a hard drive.
I'm an engineer, so I did the thing engineers do. I built something to watch for me. Not to replace being present, but to catch the part of the day that happens while a parent is cooking, on a call, or simply in the other room.
What started as a weekend project became a face-recognition pipeline, then a behavioural-analysis backend, then a phone app, then a camera of its own. It's still running in our house every day. Now we're opening it to other families.
"The cameras are always recording. Nobody has time to watch. That gap is the whole product."
Thousands of Reolink clips piling up on an external drive. No way to find a specific moment without scrubbing through every one.
A nine-phase face-recognition pipeline. First validation batch: 134 clips scanned in 3.3 minutes, 17 confident matches, every annotation correct, zero false positives.
Matched clips described by a vision-language model. Daily summaries, activity breakdowns, and developmental observations grouped by domain.
The phone app: a daily journal, clips, live view, and a library. A fine-tuned child detector took the person model from 0.33 to 0.90 mAP50 on our own footage.
Admins and viewers, invitations, multiple children per home, and the bookmark that makes a clip permanent.
10,690 clips from Feb to Jul 2026 scanned into one library. A Raspberry Pi camera kit that keeps only clips with a person in them. And a new name, courtesy of Ellie.
These aren't values on a poster. Each one is a guardrail written into the software, and each one exists because a real test exposed a gap.
Your child's face is matched on hardware you own. Biometric data is never uploaded, and only the written observation from a matched clip is stored. Video stays yours.
A false positive is worse than a miss. The pipeline is built to under-report rather than to annotate the wrong child, which is why borderline clips get a review tier instead of a guess.
PeetaBoo describes what it sees in plain language. It will never suggest that something is wrong with your child. The notes are written to be shared with your pediatrician, not to replace one.
Original footage is irreplaceable. The pipeline copies matched clips into your library and never modifies, moves, or deletes a source file.
They're the same person. That will change as the beta grows. Until then, you'll get answers from the person who wrote the code.
A parent who learned computer vision to stop missing his daughter's day. Writes the pipeline, the backend, the app, and the camera firmware. Based in Oakville, Ontario.
Almost two. Says "peetaboo". Appears in most of the footage on this site, and in every test batch since the beginning. Has strong opinions about the armchair.
We live this too. If you'd like to help shape what PeetaBoo becomes, the waitlist is the door.
Oakville, Ontario · hello@peetaboo.com