Jake Lee

About

Led architecture for “Solstice/Solstis,” a safety-aware, hands-free AI medical assistant that guides users through minor emergencies with a structured, state-machine-driven LLM agent integrated with device hardware. Built RAG grounded in Red Cross procedures plus guardrails, fallbacks, and emergency escalation, and improved real-world usability by shifting from open-ended chat to a deterministic step-by-step workflow measured via completion rate, repeat prompts, and latency.

Key Strengths

• End-to-end architecture ownership of a safety-aware LLM medical assistant agent
• Designing predictable, constrained multi-step agents (state machine vs open-ended chat)
• Separation of reasoning from execution using backend tools/functions for reliability
• Safety-first design: guardrails, fallbacks, and emergency escalation logic
• Schema/structured-output approach to avoid brittle freeform parsing
• Production evaluation and monitoring across offline tests, online metrics, and clustered error analysis
• Data normalization and validation for inconsistent semi-structured RAG sources
• Improved task completion and reduced repeated prompts by redesigning to step-based workflow
• Reduced speech-to-response latency and increased determinism to build user trust

Experience

Education

Skills

Languages