What happens when you bring a large language model to the front lines of care? This talk explores our work building a voice-driven, edge-deployed LLM system for use in the emergency department. With no internet connection, low-latency inference, and a fully local privacy-preserving design, our system runs entirely offline. It adapts dynamically to both the patient and the physician's specialty, enabling tailored language understanding in fast-moving clinical contexts. We'll unpack key design choices, challenges of real-time interaction, and what it takes to make language models clinically useful when seconds matter.

This presentation will be held in 2036 Palmer Commons. There will also be a remote viewing option via Zoom.