Pocket: an alert app that shows its work
A prototype design for a personal health alert and record companion, which you can try below as a browser-only workbench on synthetic data. In the design, your phone watches your heart-rate readings, fixed rules decide how urgent a change is and which care pathways match your record, and a clinician you choose can review the wording.
Check what an AI told you Try the Pocket workbench
What it does
- Your iPhone reads the heart-rate readings your phone or watch writes to Apple Health, alongside provider FHIR records you keep on your device and notes in your own words.
- It compares them with your own usual range at that time of day, and with the limits written in your own health record.
- Fixed rules set the urgency. A language model never sets or lowers it. The model only writes the plain wording, and that wording is checked against the readings before you see it.
- Your Mac runs the model (Gemma 4) so the phone does not have to. If the Mac is off, the phone still applies the same rules and words the alert with Apple's on-device model or, failing that, a fixed sentence.
- With no recent readings it says no recent readings. It never says everything is normal without data.
- Fixed criteria rules also check your on-device record against sourced clinical pathways, let you mark any deciding chart entry Right or Wrong, and prepare a role-filtered Omaha System summary when you share with a physical therapist, physician, nurse, or caregiver.
Interactive Pocket workbench (runs in your browser)
Every rule below runs in your browser tab on a synthetic sample record. Nothing is uploaded and no account is needed.
1. Today card — one thing you love, today, with someone
Pocket offers one daily action in your own words. Every offer has a Not now button, and logging what you did creates an on-device receipt hash without personal detail.
Today's offer in your words: "Walking the creek path with a neighbor before 9 a.m."
Waiting for your choice. Nothing leaves this browser tab.
2. Fixed urgency floor (heart-rate readings)
Switch the synthetic Apple Health reading stream. A language model cannot lower the urgency set by these rules, and zero readings never produces an all-clear.
Within your usual morning range (58 bpm against 54–62 bpm).
Fixed rule check: 7 readings in the last hour, none above your 100 bpm resting limit.
3. Your record & accuracy check (Right / Wrong)
Provider FHIR entries sit next to your own words. Notice the second entry: an imported hospital note mistakenly lists a prior knee replacement. Mark it Wrong (dispute entry) and watch the pathway below update immediately.
Deterministic criteria sets check your active record and point to the matching tool across the portfolio. Every pathway has a Not now button, and nothing is sent unless you click a door.
5. Share with a professional (Omaha System per-role cut)
When you share a summary, fixed rules cut the record by role in under a millisecond. Caregivers see what you love and movement safety notes, never medication doses, diagnosis codes, or raw blood-pressure numbers.
Role taxonomy: 374U00000X (Home Health / Personal Care)
Caregiver view — plan, safety notes, and what the person loves (no doses, no diagnosis codes, no lab values)
Stripped by rule in this view: Medication doses · ICD-10 / RxNorm / LOINC codes · Raw blood pressure and lab numbers
6. Stream-of-consciousness story (Apple HealthKit & Google Fit + Gemma 4 grounding)
While a note vault is like a book you flip through by chapter, Pocket weaves your live vitals stream, what you love doing, and your hierarchical clinical record into a continuous spoken story — checked so Gemma 4 never invents a number, a diagnosis, or false reassurance.
Stream: Apple HealthKit · Engine: Gemma 4 (on-device grounding gate)
7. Visual learning & clinical image pre-screen (missing-data gap detector)
Before a visit or home exercise check, Pocket checks your active record for missing baseline data (weight-bearing X-ray, wound or skin baseline photo, 10-meter walk cadence, or posture check) and runs a 224x224 on-device framing and lighting pre-screen. Raw photos stay on your phone; Pocket never diagnoses from an image.
On-device 224x224 quality & framing pre-screen (never diagnoses; raw photo stays on device)
X-ray (orthopedic) — Weight-bearing joint space & alignment pre-visit capture check
Missing-data check: No weight-bearing right knee radiograph linked in the last 12 months.
Record gaps surfaced on device:
8. Phone-controlled hash-locked clinical arms & on-device multimodal visual studio
From your paired phone, you can dispatch a bounded clinical arm on your M-series Mac. Before acting, the arm locks a cryptographic state fingerprint (preHash), synthesizes a 4-panel visual storyboard from your active on-device record (excluding any chart entry you marked Wrong), and seals a post-action receipt (postHash). Zero cloud egress; every output remains an unsigned draft until you or a licensed clinician approve.
9. Unsigned clinician packet file preview
When you hold an alert for clinician review, Pocket writes an unsigned packet file containing no name, no date of birth, and no contact detail. A clinician opens it at /provider in their own browser.
Where your data goes
Your readings and record stay on your phone and your Mac. Nothing is sent to us. The one thing the app fetches from the internet is a signed review record, by its receipt id, and only when you ask it to.
Asking a clinician to look
You can hold an alert for review. The app writes a packet file: the alert level, the AI draft, what the draft was written from, and a fingerprint of your record. It has no name, no contact detail and no date of birth, and the note you type is refused if it looks like it contains one. The packet is marked unsigned until a clinician signs it. You decide who receives the file.
If you are the clinician
- Open /provider and verify your NPI.
- Choose Open a packet file. The file is read in your browser and checked there. It is not uploaded. A packet that was changed after the person made it is refused.
- Read the AI draft and what it was written from. Sign or decline with your own key.
- Give the receipt id back to the person. Their app fetches the public record and checks the signature itself.
What is stored on our side: the text the clinician signs (the decision, the alert level, the packet receipt and the AI draft's wording), its hash and the signature. Not the raw readings, not the person's note. While Pocket is a prototype, only demo packets can be reviewed. What a signature proves: the key bound to that NPI signed exactly this text. It is a review of an AI draft, not care for the person and not medical advice to them.
What is not true yet
- No public release and no App Store listing.
- No clinical validation and no study. No real watch data has been run through it.
- No clinician has reviewed a real person's alert. Nothing is charged for anything here.
Questions or a clinician who wants to try the review flow: the pilot page.