ClearCare

• 

  Nurse Side

• 

  Patient Side

Inspiration

In Newark, NJ one of America's most linguistically diverse cities emergency room patients who don't speak English face a broken system. They wait hours for human interpreters, struggle with phone translation services, or simply point at their body and hope. Nurses make triage decisions on incomplete information. People die from this.

We built ClearCare because the person who doesn't speak English deserves the same shot at surviving as everyone else.

What it does

ClearCare is a Claude-powered multilingual triage assistant. When a non-English speaking patient arrives at an urgent care clinic, they open ClearCare on a tablet. Claude automatically detects their language and conducts a full structured triage interview asking about symptoms, onset, severity, and medical history entirely in the patient's native language.

The nurse sees nothing of the patient-side conversation until Claude generates a clean English summary card with triage priority (Immediate / Urgent / Non-Urgent), red flags, and a recommended action. If the patient describes life-threatening symptoms, chest pain with arm pain, difficulty breathing, stroke symptoms, an escalation alert fires on the nurse dashboard instantly, before the conversation even ends.

The nurse makes every clinical decision. ClearCare just makes sure she has the full picture.

How we built it

• Patient side: React + Vite frontend, Deepgram API for multilingual voice input
• Backend: Node.js + Express proxy server handling Claude API calls
• AI: Claude Haiku for multilingual triage conversations and structured JSON output
• Nurse dashboard: React frontend polling the backend every 2 seconds for real-time updates
• Integration: Patient app POSTs structured JSON summaries and escalation alerts to the nurse dashboard in real time across two separate machines

Challenges we faced

Getting Claude to reliably output structured JSON after a multilingual conversation, without markdown formatting, code fences, or truncation, required significant prompt engineering.

What we learned

Claude is extraordinarily capable at multilingual clinical reasoning. One model handles language detection, empathetic conversation, medical reasoning, and structured output simultaneously no translation API needed. We also learned that the hardest problems in healthcare AI aren't technical, they're about trust, validation, and knowing when to hand off to a human.

What's next

ClearCare's real deployment path is through community health clinics and urgent care centers, not hospital ERs. Phase 1 validation requires bilingual clinical reviewers for each language added.

Built With

• claude-ai
• deepgram
• express.js
• javascript
• node.js
• react
• vite