Recovery is never a solo act
In an orthopedic clinic in Taiwan, there's a scene that plays out every day: grandmother gets a knee replacement, dad has rotator cuff repair, and the son tears his ACL. All three see the same surgeon. But the person actually managing everyone's rehab? The daughter-in-law.
She has ninety minutes after work. Three family members, three sets of exercises, three different recovery timelines. She opens the app, but it shows her three separate dashboards with three separate charts. She has to log in to each account, interpret each graph, and figure out who needs her attention most.
This isn't a technology problem. It's an attention allocation problem. And no rehab app has ever tried to solve it.
What we built
The iRehab Patient App home screen now has a feature called "Ask My AI."
Tap it, and the system packages the patient's rehab progress into a structured summary — surgery type, days post-op, pain trends, exercise completion rate, how far they are from the next phase milestone. One tap to copy. Paste it into ChatGPT, Claude, or Gemini, and ask questions in plain language.
Not an AI built into iRehab. The patient's own AI.
Why we didn't build AI in
Because AI models turn over every six months.
If we hardwired GPT-4o into the app, it would be outdated by fall. If we integrated multiple model APIs, our token costs would scale linearly with patient volume — and every model update would require re-validating clinical safety.
But the deeper issue is simpler: we don't need to compete with OpenAI on chat experience.
ChatGPT is already on billions of phones. Claude reasons better for complex queries. Gemini is woven into the Google ecosystem. These teams have spent billions of dollars refining their chat interfaces and inference quality. There's no reason for us to build an inferior version from scratch.
We have exactly one job: give AI the right context.
Without context, AI can only say "some pain after surgery is normal." With context — day 42 post-op, left knee TKA, Phase 3, ROM 105 degrees (target 110), VAS spiked from 3 to 5 after wall slides — AI can say something genuinely useful.
We call this architecture BYoAI — Bring Your Own AI.
The economics of BYoAI
| Dimension | Built-in AI | BYoAI |
|---|---|---|
| LLM cost | Scales linearly with patients | Zero |
| Model upgrades | Requires re-validation each time | Automatic (handled by AI provider) |
| Vendor lock-in | Tied to a specific model | None (plain text) |
| Chat quality | Limited by our prompt engineering | World-class AI teams optimizing for you |
| Dev maintenance | Chat UI + safety audits + model switching | One API endpoint + one copy button |
The core insight behind BYoAI: AI inference is commoditized; clinical context is scarce. Anyone can use ChatGPT. But only iRehab knows what day post-op you're on, your current ROM, and your exercise completion rate this week.
From individual to family
Once we shipped this feature, a bigger picture emerged.
That daughter-in-law can now open each family member's app, copy three summaries, paste them all into one AI conversation, and ask: "Of these three people, who changed the most this week? Who can exercise independently?"
The AI doesn't need to interpret three separate charts. It can say directly: "Grandmother's completion rate dropped 30 percentage points but her pain didn't worsen — likely a motivation issue. Dad and son are both trending stable. Suggest spending tonight with grandmother."
This is family-level rehab coordination. Not three independent logs, but a caregiver's view with clear priorities.
No rehab app has done this before. Not because it's technically hard, but because everyone treats rehabilitation as an individual activity. In Taiwan — and across most of Asia — recovery is a family affair. After surgery, the family provides care. Whether the patient actually follows through on exercises is often decided by family dynamics, not individual willpower.
Safety boundaries
We made several deliberate design choices to keep this safe:
Exports contain no personal data. No name, no national ID, no phone number, no physician name. Only clinical metrics.
Verb constraints in contextual prompts. The four default prompts use only "organize," "identify," and "list" — never "recommend," "evaluate," or "diagnose." AI is a secretary, not a doctor.
Every export carries a disclaimer. "This summary is intended for organizing questions and discussing them at your next appointment. It should not be used independently to modify your treatment plan."
Trend labels are computed server-side. "Improving," "stable," "fluctuating," and "worsening" are not generated by AI. They're calculated by our algorithm using explicit rules. Deterministic logic, not LLM speculation.
Next step: Family Link
Today, caregivers need to log into separate accounts to copy each summary. We're building Family Link — a feature that lets a caregiver link family members within their own account and view all summaries in one place.
Linking requires three-layer verification: national ID (identification) + date of birth (proof of relationship) + mutual phone confirmation (proof of account access). And you can only link patients who see the same physician — not arbitrary users in the system.
Family Link access is scoped to AI Prep summaries — read-only, with no ability to report exercises or modify any data on behalf of family members. Compared to today's reality of shared passwords, this is actually more secure: narrower scope, full audit trail, and either party can unlink at any time.
The real product category
We started building "post-surgical rehab tracking." But as the product evolved, we realized we're actually building family-centered rehab coordination.
Individual tracking is the entry point. Family coordination is the moat.
In orthopedics, family-cluster referrals are the most natural growth loop: grandmother gets a knee replacement and refers her daughter for a shoulder issue, the daughter refers her husband for ACL reconstruction. This acquisition pattern is organic. But to turn it into a product advantage, your system has to understand the concept of "family."
Most rehab apps don't. They assume each account is an isolated individual.
We don't.
iRehab turns daily rehab data into actionable context for patients, families, and clinicians. From individual tracking to family-centered recovery coordination.