Launching Vertia.
One number a day. A coach who actually knows your body. The thing I wished existed when my physio first told me my limp was protecting something I couldn't feel.
I built Vertia because no app I could buy did the one thing I wanted: read my body, daily, with zero ceremony, and tell me when I was drifting before I felt it.
The whole thing in one sentence.
Pop in your AirPods. Stand still for 30 seconds. You get one number, 0 to 100. That's the morning ritual. Lower means more drift, higher means steadier. Trend visible on day three. That's the whole product, and everything else is in service of that number — the walking coaching, the conversational coach, the photo scans, the body silhouette, the clinical ingest.
The three rituals.
Morning — the 30-second daily habit loop. AirPods stream head pitch/roll/yaw at 25 Hz. We extract sway magnitude, roll bias, pitch bias, breathing rhythm, jaw clench baseline. You see the score, not the spreadsheet.
Walking — phone in pocket, AirPods in ears. Vertia auto-detects the walk via CMMotionActivity, streams head IMU plus step timing from the iPhone pedometer, and when your step-interval asymmetry crosses 8% with a 30+ step sample, the coach speaks one spatial cue in the heavy side's ear. Rate-limited to one cue every 90 seconds. Never preachy.
Tell-it — the composer at the bottom of the home screen accepts speech, the camera, photos, files. You can drop an X-ray, an MRI report, a scribbled note from your physio, or a 30-second voice memo about why your back hurts today. Claude reads it, returns a JSON of body map updates plus medical history entries, and the body silhouette updates region by region with a cascade animation.
The coach.
The coach has three personalities — Bro, Mentor, Drill Coach — each paired with a default voice but independently swappable across five ElevenLabs voices. Every reply is grounded in your actual data: latest morning checks, recent test results, medical history pulled by user_id, and a live sensor snapshot you send up with each message. The system prompt forbids inventing numbers and limits to exactly one action per turn.
For walking, coaching is deterministic, not LLM. The threshold engine evaluates sensor expressions and picks the highest-priority cue not recently used. LLMs are too slow and too unpredictable for real-time biomechanical feedback — they're for the explainer, not the live ear.
What it sees.
iPhone CMPedometer for steps, cadence, distance. iPhone CMMotionActivity for walk/run/stationary classification. AirPods CMHeadphoneMotionManager for head pitch/roll/yaw and inward-mic audio at 16kHz+. iPhone Vision framework for body pose extraction on the photo scans — 17 3D joints, 76 face landmarks. HealthKit for walking asymmetry %, walking speed, step length, double support, stair speed.
Raw sensor data stays on device. SwiftData stores everything; only aggregates and derived signals sync to Supabase. You own the recording.
What's next.
Walk-by visual scans (full gait cycle from a propped phone), rotation scans (360° in front of the camera), watch companion (arm swing + haptic correction cues), the clinical PDF generator so you can hand a physio a real artifact, weekly Claude digest with the narrative.
One more thing.
If this resonates and you have AirPods, I'd love you on the TestFlight. The app is iOS 26 only — the new Liquid Glass APIs do a lot of the visual lifting and they're not back-portable. Email me at iliescualex@yahoo.com with your Apple ID and I'll add you.
Alex Iliescu — Bucharest, May 28 2026.