iOS · COMPUTER VISION · 2023 to 2024

PhysioRise

An iOS fitness app that generates personalized warm-ups and corrects exercise form in real time using on-device pose estimation. Sub-second feedback loop, no network call, fully private. Plus a doctor portal so PTs can prescribe routines and see progress.

SwiftUICoreMLVision AVFoundationHealthKit Synopsys SCVSEF · 1st Place CSEF Qualifier
17
Body keypoints tracked
30 fps
Live pose inference
100%
On-device, no cloud
2
Apps shipped (patient + dashboard)
PhysioRise running live: skeleton overlay classifying a squat at 99.2 percent confidence on the left, and a Low Confidence / Observing state on the right when no exercise is being performed
FIG 01 · Live in-app classification · skeleton overlay + per-exercise confidence iPhone · CoreML on-device
01 · PROBLEM

Bad form is the hidden cost of telehealth physical therapy.

Patients prescribed home exercises don't have a clinician watching. Form drifts. Drift causes reinjury. Reinjury sends them back to the PT, often weeks later, and the relationship between effort and recovery breaks.

~50%
Of physical therapy patients abandon home exercise programs within 4 weeks.
2-3×
Higher reinjury risk when exercises are performed with poor form.
$0
Real-time feedback offered by a paper exercise sheet.
02 · SYSTEM

Two apps, one ML pipeline.

PhysioRise is a SwiftUI app with a CoreML pose model running through Apple's Vision framework. The companion PhysioRise Dashboard is the clinician-facing app where PTs prescribe and review.

Capture
AVFoundation
Front camera · 30 fps
Detect
VNDetectHumanBodyPose
17 keypoints, on-device
Classify
PT_Model_InfinityAI.mlmodel
Action classification per exercise
Coach
SwiftUI overlay
Skeleton + form cue
PhysioRise training pipeline: synthetic exercise data on the left, Create ML Action Classifier UI on the right showing 88.8 percent training accuracy and 79.4 percent validation accuracy across 220 iterations
FIG 02 · Training data + Create ML Action Classifier 88.8% train · 79.4% validation · iter 220
03 · COREML + VISION

A custom Action Classifier per exercise.

For each prescribed exercise (squat, lunge, glute bridge, plank, etc.) PhysioRise loads a different Create ML Action Classifier. The model takes a sliding window of pose frames and outputs both the recognized exercise and a confidence-weighted form score.

Why on-device

PT data is health data. Streaming video to a server requires HIPAA infrastructure that an indie iOS app cannot support. Apple's Vision + CoreML stack runs the entire pipeline on the Neural Engine. Inference is free, private, and offline-capable.

Why per-exercise models

A single multi-class model gets confused between similar movements (lunge vs. split squat). Per-exercise classifiers are smaller, faster, and easier to retrain when a PT requests a new prescription type.

04 · DOCTOR PORTAL

Where the PT actually closes the loop.

A separate macOS-style dashboard (PhysioRise_Dashboard) where licensed PTs prescribe exercises, set rep targets, and review session-level adherence and form scores. Patients see their progress; PTs see the cohort.

SurfaceAudienceStack
PhysioRise (patient)End user, homeSwiftUI · Vision · CoreML · AVFoundation
PhysioRise DashboardPhysical therapistSwiftUI · CloudKit sync
Physiorise_patientPatient (companion)SwiftUI · onboarding flow
physiorise-therapist-appCross-platform PTReact Native · Expo
05 · RESULTS

Recognition.

Outdoor testing photos showing PhysioRise classifying pushups, squats, and superman exercises with green borders for correct classifications and red borders for those needing correction
FIG 03 · Field validation · green = correctly classified, red = flagged for review iPhone outdoor session
1st
Synopsys SCVSEF, Castro Family Award for Best Physical Science Project.
CSEF
California State Science and Engineering Fair qualifier.
JIC
Nominated for the Thermo Fisher Junior Innovators Challenge.
5
Iterations of the Xcode project (Physiorise_patient through PhysioRiseeee).