AI / Healthcare
LLM Features in Clinical Workflows
Patient chat, triage assistance, and clinical note drafting integrated into backend workflows, with consent checks and outage fallbacks.
- Node.js
- LLM / RAG services
- Twilio Voice
- MongoDB
The problem
AI chat and triage carried real patient-safety edge cases, such as an opted-out patient receiving an AI reply or a clinical note drafted without human review.
What I built
An isolated LLM service boundary with consent and tenant checks ahead of every call, defined fallbacks for a degraded service, and per-hospital enablement.
My role
- Integrated RAG chat, triage, and note drafting
- Consent and tenant checks before every LLM call
- Fallback behavior for LLM outages
- Voice AI agent integration
Technical highlights
- AI features can be enabled, disabled, or rolled back per hospital
- A human review step for drafted clinical notes
- Defined behavior when the LLM service is degraded or unavailable