Agentic voice AI that calls discharged patients in their own language and walks them through recovery, built to cut avoidable readmissions.
Top 5 at HackUSF 2026. Built in 24 hours with Alan Risal, Kevin Fernandez, and Sharan Santharam.
1 in 5 patients are readmitted within 30 days. Often it is not a medical failure: they were discharged exhausted, handed a six-page document they could not fully process, and then left to manage recovery alone.
CareCall calls patients the day after discharge, in their own language, and walks them through their recovery plan.
- A 9-step care-protocol state machine guides each call through the recovery instructions.
- Auto-escalation to nurses the moment an urgent flag is raised.
- A structured post-call report is generated and emailed to the physician after every call.
- Voice in 70+ languages via ElevenLabs.
A multi-agent system on Google ADK + Gemini. Gemini handles call summarization and highlight extraction, turning each conversation into a structured follow-up document for the doctor. ElevenLabs provides natural multilingual voice. A HIPAA-compliant Supabase backend stores encrypted PHI.
Discharge -> Voice call -> Comprehension check -> Triage / escalate -> Physician report
| Layer | Tool |
|---|---|
| Agents | Google ADK |
| Reasoning | Gemini |
| Voice | ElevenLabs (multilingual, 70+ languages) |
| Backend | Supabase (HIPAA-compliant, encrypted PHI) |
| Frontend | React |
I built the multi-agent architecture: the Gemini-powered call summarization and highlight extraction that produces the structured physician report after each call, plus the ElevenLabs voice integration across 70+ languages.
Team project built at HackUSF 2026 with Alan Risal, Kevin Fernandez, and Sharan Santharam. Originally authored as DischargeGuard by @alanrisal; this repository is the same project.