How AI + Hardware Can Transform Point of Care Workflows
Presents a pocket-sized 10-watt bedside inference device running local 3B language models on a CM5 and RP2040 for private clinical workflows.
Summary
Clinical Workflow Bottlenecks in Healthcare
Healthcare environments generate massive volumes of patient vital signs, medical imagery, and clinical notes. However, accessing and processing this data at the bedside is constrained by strict privacy regulations and unreliable hospital Wi-Fi networks.
Custom Embedded Hardware Architecture
PamirAI engineers developed a compact, 10-watt bedside inference device incorporating a Raspberry Pi Compute Module 5 (CM5), an RP2040 co-processor, and a sunlight-readable E-ink display tailored for clinical environments.
On-Device Local Inference
The device executes a 3-billion-parameter language model entirely on-device alongside local speech-to-text transcription models, providing private, low-latency AI reasoning without transmitting clinical data to cloud servers.
Offline Sensor Integration and MCP Hub
An embedded software hub interprets natural language instructions (such as 'log heart rate every five minutes') into automated sensor data collection routines, enabling rapid device deployment even during network outages.
Related talks
Creator Stage·13 min
Kevin · Tianqi
Demonstrates custom CM5 and RP2040 embedded hardware paired with local speech and language models for automated offline clinical data logging.
Creator Stage·22 min
Dr Katie Paxton Fear
Demonstrates using AI-assisted coding to reverse engineer conventional hardware and integrate legacy sensors into Home Assistant.
Village Interviews·9 min
The DEF CON Video Team gets the full overview of hands-on hardware labs, firmware analysis, and microelectronics at the Embedded Systems Village.