GoldenHour AI: Automated Bed Routing

PriorityPulse eliminates bottlenecks by combining LLM chart parsing with live FHIR logistics. Our A2A system automates triage and bed matching to save the Golden Hour.

Comment

Inspiration

The 2026 Hospital Crisis isn’t just about a lack of beds; it’s about a lack of time. Clinicians are experiencing unprecedented burnout because they spend hours manually reviewing fragmented patient charts, calculating risk scores, and calling bed managers to find available rooms. When a patient is experiencing a time-critical emergency like a STEMI or a stroke, "time is tissue." I was inspired to eliminate the cognitive load and logistical bottlenecks of hospital handoffs by building a system that lets doctors focus on medicine, while the AI handles the logistics.

What it does

GoldenHour AI: Automated Bed Routing is an intelligent Agent-to-Agent (A2A) clinical triage system. It transforms messy, unstructured clinical handoff notes into actionable logistical commands in seconds.

Unstructured Ingestion: A triage nurse simply uploads a standard patient clinical record.

Clinical Intelligence: The agent parses the data, extracts the primary diagnosis (e.g., STEMI), and performs real-time clinical math to calculate the patient's "Golden Hour" remaining window (e.g., the 90-minute door-to-balloon PCI window).

Live FHIR Bed Routing: Instead of hallucinating a room assignment, our agent securely reaches out across the internet to the hospital's live FHIR database, queries the Location resources, filters out occupied beds, and matches the patient to the first available, Tier-appropriate bed (e.g., routing a Tier 1 patient to an unoccupied Resuscitation Bay).

How we built it

I built My solution using a robust, enterprise-grade architecture:

The Brain: I utilized the Prompt Opinion platform and its native LLM orchestrator to handle the initial unstructured document reading and user intent recognition.

The Engine: I have customized a prompt opinion github repositorie to add my tools inside and logic.

The Tools: My architecture relies on a strict division of labor between the cloud orchestrator and my custom-engineered clinical microservice

The Bridge: I hosted my local Express server and tunneled it to the cloud using ngrok, and after for marketplace i used render.comto deploy my srver establishing a secure Agent-to-Agent (A2A) JSON-RPC connection where the platform orchestrator could dynamically hand off tasks to our specialized clinical agent.

Challenges we ran into

Building a true multi-agent system is complex, and i hit several major architectural hurdles: LLM Tool Hallucinations: I had to strictly prompt-engineer my agent so it would physically execute the getAvailableRooms database query rather than just guessing or making up an empty bed based on the text file.

Accomplishments that we're proud of

True Agent-to-Agent Orchestration: I successfully got two separate AI entities talking to each other. The platform agent reads the file, and my custom TypeScript agent executes the live database logistics.

Real-Time Clinical Math: I didn't just build a text summarizer; i built a system that actively compares symptom onset timestamps against current times to calculate dynamic "Golden Hour" countdowns.

Zero-Click Logistics: I completely removed the manual step of checking hospital census data. My agent successfully bypasses occupied rooms and finds the needle in the haystack (an empty, Tier-1 capable bed) instantly.

What we learned

I learned an incredible amount about the future of AI in healthcare. Specifically, I learned how to implement the A2A (Agent-to-Agent) protocol to build modular AI microservices. I learned how to interface with FHIR R4 standard databases, and i gained deep experience in strict TypeScript enforcement, Express server management, and advanced prompt engineering to constrain LLM behavior to factual, tool-based outputs.

What's next for GoldenHour AI: Automated Bed Routing

This hackathon prototype is just the beginning. My roadmap for solving the 2026 crisis includes:

Specialized Downstream Agents: Routing the output of this agent directly to a "Pharmacy Agent" to automatically prep medications (like Heparin) before the patient even arrives at the bed.

EHR Write-Back: Allowing the agent to automatically document the triage scores and bed assignments back into the patient's official FHIR record.

Push Notifications: Integrating the SDK's push notification capabilities so the assigned Care Team receives an immediate alert on their devices the second the patient is routed to their bay.

Built With

Share this project:

Updates