Inspiration
Hospitals lose millions due to "Invisible Gaps"—exam rooms that sit empty while patients wait, simply because scheduling systems aren't synced with physical reality. Inspired by Operational Excellence, I built HospitalOS to bridge this gap using Ambient Intelligence, turning sound into actionable data without compromising privacy.
What it does
HospitalOS is an AI-driven orchestrator that monitors hospital environments (OR, ER, Exam Rooms) via ambient sound. Real-time Detection: Identifies machine idle sounds or room silence using Google YAMNet.
Reasoning: Uses Gemini 2.0 Flash and GraphRAG to compare physical status with scheduling systems.
Automated Action: Updates room status, calls patients, and optimizes billing automatically when a "Scheduling-Reality Mismatch" is detected.
How we built it
I developed a multi-layered pipeline for safety and precision: Audio Pipeline: Integrated Silero VAD for privacy masking and YAMNet for acoustic labeling.
Reasoning Engine: Built a GraphRAG framework to handle complex medical workflows and conflict resolution.
Backend: Connected a Python AI pipeline with a Spring Boot server to manage secure command execution.
Challenges we ran into
Processing 32-channel audio in real-time while maintaining 100% privacy was the biggest hurdle. I solved this by implementing a Privacy-First VAD filter that masks human speech before it reaches the reasoning engine.
Accomplishments that we're proud of
The AI Thought Process (Decision Timeline) UI. It transforms "black box" AI into a transparent agent by showing exactly which signals led to an action, building trust with medical staff.
What we learned
I discovered that Ambient Sound often carries more operational context than visual data in a hospital setting. I also learned how to effectively orchestrate LLMs with Structured Ontologies to prevent hallucinations in mission-critical environments.
What's next for HospitalOS: AI Ambient Orchestrator
Multi-modal Expansion: Integrating low-cost LiDAR to complement audio data for higher precision in crowded ERs. Predictive Analytics: Using historical "gap" data to predict bottleneck periods before they happen. Global Compliance: Adapting the engine for GDPR/HIPAA certified cloud regions to scale into European and US markets.
Log in or sign up for Devpost to join the conversation.