Overview
AI engineer and full-stack developer at KPMG, building a patient-facing AI avatar for a hospital deployment. The goal is to absorb routine patient questions (where do I go, what's the prep for tomorrow's procedure, what is this medication called and why am I taking it) so nursing staff can focus on the parts of patient care that actually need a nurse.
Client name confidential.
What I own
- Agent design. Designing the AI agents on Microsoft Foundry: the orchestration of skills, retrieval, and conversational flow that produce safe, in-domain answers. "Safe and in-domain" is the harder of the two. In a hospital setting, the failure mode of a confidently wrong answer is much worse than a polite "I'll get a nurse for you," and the agent has to know the difference.
- Frontend. The React application that serves as the avatar's interface to the patient.
- Glue. Full-stack development connecting the AI backend to the patient-facing frontend, including session and state handling.
- Healthcare context handling. Making sure sensitive contexts (medication queries, distressed patients, anything that could be construed as medical advice) route to a human or to a constrained safe response, not to the LLM's open generation.
Stack
- AI orchestration: Microsoft Foundry (agents, retrieval, evaluation)
- Frontend: React
- Role: AI Engineer + Full-Stack Developer
What I'd add to this page later
Same constraint as the Health Platform page: client-confidential. Non-confidential additions worth surfacing (the size and shape of the agent, the evaluation framework, the patient-flow categories the avatar handles) are coming once I know what's safely shareable.
