A running, queryable model of your health — built from wearables, labs, and your own words — that prepares you for the doctor instead of trying to replace one.
Sparked by Reid Hoffman's "digital twin" idea for personal health (Axios, 2024): most people's health data is scattered — a portal here, a wearable app there, a paper printout in a drawer — and nobody is reading all of it together except you, badly, at 2am. A twin is one continuously updated model that can read all of it, and tell you what's worth asking about.
What it would do
1
Ingest what already exists
Wearable exports, lab PDFs, pharmacy records, your own symptom notes — pulled in with your consent, not scraped. No new sensors required to start.
2
Hold one model, not a pile of files
Trends and correlations over time (resting heart rate drifting, a lab value creeping toward a range, a pattern between sleep and mood) — the thing a single 15-minute appointment can't see.
3
Prepare you, not diagnose you
Turns "what should I ask my doctor" into an actual short list, in plain language — the twin's job ends at the clinic door.
⛔
No raw health data to a rented model — ever
The same PII-firewall posture as the rest of this estate, but for the most sensitive category there is: tokenize before anything leaves your tenant, watcher blocks raw egress, fail closed.
A worked sketch
Illustrative only — no such pipeline exists yet. Names and numbers below are invented.
Twin noticesResting HR +6bpm over 8 weeks; sleep down ~40min/night; last lab flagged borderline ferritin.
What it tells youThree things worth a 5-minute mention at your next visit — not a diagnosis, a shortlist.
What it never doesName a condition, adjust a medication, or replace the clinician's read of you.
⚖️ Honest scope — what this is and isn't
Nothing here is built. No ingestion, no storage, no model, no UI. This page is a concept sketch, in the same register as CongoSky's other vision-only workstreams (Migration/Emancipation, CongoPhone) — named, not faked.
Health data is the highest bar this estate has. It would need everything the PII firewall already proves for identity data, then more: real clinical review before any output resembling advice, and no "AI diagnosed me" framing — ever. This estate's own honesty audit already flags that no specialist review has happened on other health-adjacent work (Ikhaya/Qapha); the same bar would apply here, harder.
A companion for the visit, not a substitute for the clinician. Its entire value is a better 15 minutes with a real doctor — not a verdict delivered instead of one.
Would reuse this estate's existing patterns, not invent new ones: the PII firewall for tokenize-before-model, the one-record/delta-store for an append-only personal history, and the four-gate policeman for human confirmation before anything is written or shared.