My Pixel had 8,743 photos and nobody was ever going to sort them. So I built Nytcrew (said "night crew"). It cleans up your gallery while your phone charges overnight, a fit-n-forget kinda solution.
The one rule: photos never leave the phone.
- Describe on device. An edge model runs locally on your phone and decides if an image is useful and reviews it with you if it feels otherwise.
- Mask, then score. Emails, links, @handles and long digit runs are masked. Only that text goes to a Cloud Function, and the same text gets repurposed for image search!
- Learn. Every keep or delete trains your model to be better at guessing your preference.
The tool has gone through many changes in its lifecycle. From being a complete cloud model using gemini to now a privacy first local model, there are exciting use-cases ahead that help us finally leverage the surprising amount of computing power that most of us overlook on our phones. Now the focus is on transferring the cognitive load of decluttering my phone, soon we will be working on tools that help us delegate many more activities that feel like chores.
The hard part was physics, not AI. One photo takes about 65 s on a Pixel 8a, so scanning runs in cycles: HeatGuard stops when the phone gets warm, rests 10 minutes, then goes again. But you would be surprised at how underutilised your phone is, given the compute power it beholds!
Top comments (0)