Inspiration
This started personal.
I've spent years navigating a healthcare system that wasn't built for people like me — years of back and forth before a proper diagnosis, sitting in waiting rooms realizing I'd forgotten the one thing I meant to bring up, being asked for the fifth time what medications I'm on. And then I became a caregiver. I watched someone I love struggle to find words for what they were experiencing, leave appointments without the follow-up they needed, and carry the weight of a fragmented system with no support.
Then we read Dario Amodei's Machines of Loving Grace.
In it, he writes that most people have never had access to a brilliant friend who happens to have the knowledge of a doctor — someone who knows their full history, speaks frankly, and actually engages with their situation. That sentence stopped us cold. Because that's exactly what we'd been missing. Not better technology. Not more portals. A friend who knows your story and helps you navigate.
Amodei's essay is optimistic about what AI could do for human health — but he's clear that the gains only materialize if the human coordination layer keeps up. Right now, it doesn't. Patients are showing up to appointments as the weakest link in their own care chain, carrying fragmented histories in their heads, missing the patterns that matter, leaving without the follow-up they need.
Lumio is infrastructure for that coordination layer. Not just a visit prep app — a foundation for what it looks like when every patient finally has that brilliant friend in their corner.
The numbers reflect what we lived:
- $25B+ wasted annually on redundant testing (National Academy of Medicine)
- $30% of lab tests ordered are duplicates of prior results (GAO)
- $15 min average appointment — most spent catching up, not caring
Lumio didn't come from a market analysis. It came from too many waiting rooms, too much history carried in our heads, and wishing someone had helped us show up prepared.
What it does
Your health data exists. It's scattered across portals, buried in PDFs, locked in systems that don't talk to each other — and the burden of connecting it all falls entirely on you. Lumio changes that.
At its core, Lumio is about ownership. Your records, your history, your story — centralized, organized, and in your hands. Not your insurance company's. Not your hospital's. Yours.
But ownership alone isn't enough. Data sitting in one place is still just data. What makes Lumio different is what happens next — the AI reads across your entire history and finds what you couldn't see yourself.
Not just "your Vitamin D is low" — but "your Vitamin D has been declining for 4 consecutive months despite supplementation, your sleep quality dropped 35% over the same period, and you've logged joint pain 10 times in the last 60 days — and you've never mentioned any of this to your doctor." Those aren't three separate data points. They're a pattern. And patterns are what get missed.
This is the vision Dario Amodei describes in Machines of Loving Grace — a brilliant friend who has the knowledge of a doctor, knows your full history, and actually engages with your situation. Most people have never had access to that. Lumio is built to change that.
The AI doesn't replace your doctor. It makes sure your whole story is in the room when you see them.
Before your appointment, Lumio ingests your health profile — medications, conditions, allergies, family history, self-reported symptoms, lab results, and wearable data — and runs it through a pattern detection engine powered by the Claude API. It looks for trends in lab values moving in a direction over time, not just whether they're in range today. It finds correlations you wouldn't catch — sleep quality dropping after a medication change, fatigue tracking with thyroid numbers, a supplement that isn't working. It surfaces gaps — symptoms you've logged but never mentioned to a provider. It flags medication interactions, values outside range for the first time, and patterns that warrant clinical attention. The output is a prioritized visit prep card: what's changed since your last appointment, what to flag, and a generated question list tailored to your visit type. The most important things surface first — because 15 minutes goes fast.
During your appointment, one tap starts a recording with real-time transcription via Whisper API. No setup. No friction. Nothing missed.
After your appointment, Lumio generates a clean visit summary within minutes: what was discussed, what was decided, and a list of action items with due dates — shareable with a caregiver, partner, or the next provider you see. The new data feeds back into the pattern engine. Every visit makes the next prep smarter. The pre-visit flags get compared against what the doctor actually addressed, so nothing falls through the cracks across appointments.
Most tools store your data. Lumio reads it — across time, across data types, across the gaps between appointments. The Claude API reasons over your full health history the way a doctor would if they had unlimited time to review your chart before every visit. It contextualizes findings, not just surfaces them. Not "this is abnormal" but "here's why this might matter, and here's what to ask."
The AI knows its lane. Every insight is framed as something to bring to your doctor — never a diagnosis, never a treatment recommendation. The goal is to make you a better-prepared partner in your own care. Your doctor remains the expert. Lumio makes sure you show up worthy of the time you have with them.
You own your records. The AI finds the signal. You walk in prepared.
How we built it
| Layer | Technology |
|---|---|
| Frontend | React Native — mobile-first, card-based |
| Design System | Claude Design |
| AI / Pattern Detection | Claude API - Opus and Sonnet |
| Voice transcription | 4o Mini Transcribe |
| Data | Local JSON profiles — no backend, on-device |
The core of the build is the pattern detection prompt: getting Claude to reason across labs, symptoms, medications, and wearables simultaneously and surface clinically meaningful correlations in plain language — without crossing into diagnosis territory.
For the demo, we seeded a realistic $5-month patient profile (Courtney H., Hashimoto's/autoimmune) with correlated signals across labs, wearables, and symptom logs — enough history for the engine to find signals the patient had never connected herself.
Challenges we ran into
We're patients and caregivers, not clinicians. Our credential is lived experience — which is exactly what drove us to build this, and exactly what made it hard. We made judgment calls about clinical framing without a medical expert in the room. Where is the line between a pattern worth surfacing and one that causes unnecessary alarm? We don't have a formula for that. We made our best calls.
We worked with mocked data. Real health records are messier, more incomplete, and harder to reason over than any seed file we could build in a hackathon. How does the pattern engine perform when inputs have missing date ranges, inconsistent lab formats, or years of sporadic logs? That's still an open question.
HIPAA got personal fast. When you're building something that holds medication histories, symptom logs, and visit recordings, the weight of that responsibility becomes very concrete. Every design decision ran through a different filter. It made viscerally clear how much — legal, technical, and ethical — sits between a hackathon demo and something you'd trust with your real health data.
The clinical boundary is harder than it looks. Early AI outputs read like a differential diagnosis. Getting the framing right — useful without being alarming, specific without being prescriptive — required more iteration than anything else in the build.
Accomplishments that we're proud of
The demo moment works. When Courtney's prep card loads and surfaces five findings she had the data for but had never connected — including a potential early autoimmune signal her doctor hadn't heard about — it feels like something genuinely new.
We're proud that the ethical constraints were first-class decisions, not afterthoughts. Consent by default. Patient-owned data. Explicit AI labeling. A hard clinical boundary. These aren't features — they're the foundation.
And we're proud that this exists at all. We're people who had a problem, learned enough to build something, and shipped it. That matters.
What we learned
The problem is real and urgent. Patients are the weakest link in their own care chain — not because they're failing, but because the system was never designed to support them.
Trust is the hardest thing to design. In health, every word and every UX choice either builds trust or erodes it. Calm language, progressive disclosure, honest AI labeling — these matter more here than in almost any other product category.
Lived experience is a legitimate foundation. You don't need to be an expert to recognize a broken system. Sometimes you just need to have lived inside it long enough to know exactly what's missing.
AI can be your whole product team. We used Claude end-to-end — not just as the intelligence inside Lumio, but as the tool we built Lumio with. Claude Design helped us go from concept to designed screens without a dedicated designer. Claude helped us write the pattern detection prompts, draft the research, pressure-test the clinical framing, and generate the seed data that powers the demo. The same technology we were building for patients was the technology building with us — and that recursive loop taught us something real about where AI-assisted product development is headed. A small team with a real problem and the right tools can move faster than we ever thought possible.
What's next for Lumio
The hack runs on local mock data. The real version needs:
- Real data ingestion — Blue Button FHIR for labs and claims, Apple Health for wearables
- HIPAA-compliant infrastructure — AWS GovCloud, encryption at rest and in transit, patient-controlled architecture
- Clinical validation — a medical advisor to pressure-test the pattern engine's output and framing
- Expanded pattern detection — more signal types, longer time horizons, condition-specific reasoning
- Provider sharing — one-click summary export formatted for clinical handoff
The longer-term vision: Lumio as the infrastructure layer for patient-side care navigation — the thing that makes every appointment, every referral, and every new diagnosis less overwhelming for the people living through it.
Your health, illuminated.
Built With
- claude
- next.js
- supabase
- vercel
Log in or sign up for Devpost to join the conversation.