Problem
Desk workers often know they should stretch, but the moment of action gets blocked by small frictions: deciding what to stretch, figuring out how long to spend, and remembering what they already did recently.
Project
Desk Flex
A local-first wellness app for desk workers that recommends what to stretch based on body-area priority, recent history, and what has already happened in the current session.
Product Approach
The core of Desk Flex is a lightweight recommendation model. It ranks stretches by urgency, using recent history and body-area weights for common desk-worker issues like hip flexors, chest, neck, lower back, shoulders, and thoracic spine.
That recommendation logic is shaped by a few practical rules. If you have already stretched one body area several times in a row, the app rotates to another area that still needs attention. If a stretch has a left and right side, it only counts as complete after both sides are done. And if you skip something, the app distinguishes between "not now," "I already did this," and "do not suggest this again," so it learns without turning a skip into bad history.
On the product side, Desk Flex treats stretching as opportunistic instead of scheduled. The app opens to one clear decision: start stretching. Sessions can last one stretch or many, and the product handles prioritization and memory in the background.
A few details make it feel more intentional than generic: all user data stays local, real exercise photos appear only where there is a strong public-domain match, and the underlying logic is covered by focused unit tests for ranking, hidden stretches, rotation, skip behavior, and body-area summaries.
Screenshots
Where It Could Go
The natural next step is an evidence-grounded coaching layer: intake and constraint capture, recommendations sourced from public guidance rather than model priors, red-flag and contraindication handling, and a transparent eval harness for the kinds of edge cases that make health-adjacent AI either useful or dangerous.