# HomeReady

- **Event:** [The Future of Agentic AI in Healthcare - Abridge x Anthropic x Lightspeed](https://cerebralvalley.ai/e/abridge-hackathon)
- **When:** Sat, Jul 18 at 9:00 AM – 10:00 PM (PDT)
- **Where:** San Francisco, CA
- **Placement:** 1st Place
- **Team:** [Pranay Madan](https://cerebralvalley.ai/u/pistachio_pranay)
- **GitHub:** https://github.com/pistachiopranay/homeready-abridge-hackathon
- **Demo video:** https://www.youtube.com/watch?v=EY5Lbxpk-V8
- **Gallery:** https://cerebralvalley.ai/e/abridge-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/abridge-hackathon/hackathon/gallery/23

HomeReady -  verify the home before discharge.

The problem
Nearly every discharge plan ends with some version of "discharge home once
safe." But nobody can verify "safe" from the chart. Post-discharge falls are a
leading driver of readmissions, and home-safety evaluations (CDC STEADI, HSSAT)
require a clinician home visit that almost never happens. The chart stops at the
encounter — the home it's sending the patient back to is a blind spot.

What it does
HomeReady turns the Abridge care plan into a guided home walkthrough. A
caregiver walks the patient's home with an Iphone app, and an agent 
runs the visit: reading the encounter note, deciding what to verify, asking
adaptive questions, and grading the home against the actual discharge plan.

It's not a scripted checklist. Grounded in Monica's record, HomeReady decides
what matters for her — osteoporosis, a prescribed walker, mobility — and the
decisive finding is a measurement, not a vibe: the bathroom doorway is too
narrow for her walker, so the discharge plan cannot work as written. That
evidence travels back upstream and coordinates a fix — routing findings to OT,
coverage barriers to social work, and orders to the care team — then updates the
chart for the final discharge decision.

How it's built
"Dumb client, smart backend." The iPad only scans and streams; all intelligence
runs in a FastAPI backend. Three perception layers run live during the
walkthrough: RoomPlan LiDAR for exact geometry (real walker-clearance math), a
Claude vision fast-pass on camera frames feeding the voice agent, and a deeper
Claude pass grading STEADI/HSSAT findings. The ElevenLabs voice agent's LLM *is*
the backend — one brain sees the chart and the camera and writes Riley's next
sentence. Findings write back as draft FHIR Observations, ServiceRequests (DME
with Medicare coverage flags), and Tasks.

What's real
Every model call in the demo happens live. There is deliberately no fabricated
risk score — findings are graded against CDC STEADI / HSSAT with per-patient
rationale, and every order and escalation is drafted, never auto-sent. Monica is
a synthetic patient from Abridge's ambient-FHIR dataset.

Abridge captures the encounter. HomeReady verifies the home.

---

Markdown version of https://cerebralvalley.ai/e/abridge-hackathon/hackathon/gallery/23. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt