ERgent | Devpost

Inspiration

This project was inspired by a very personal experience. One of our teammates had to go to the ER twice in a week with chest pain and shortness of breath. It was a frightening moment for all of us. Once we arrived, we noticed how inefficient the triage process felt: patients with potentially life-threatening conditions sat waiting in a crowded room while overwhelmed staff struggled to keep up. We wanted to build a system that makes emergency triage more efficient and ensures that patients with severe symptoms are prioritized. Being in the ER should bring a sense of safety — not heighten anxiety.

What it does

ERgent is a multi-agent AI triage support system that integrates into existing emergency department workflows. The platform retrieves relevant patient history from stored records, assigns severity levels using Emergency Severity Index (ESI) principles, and generates guideline-based recommendations for clinicians. It also analyzes radiology findings in the context of each patient’s history, manages patient queues with real-time updates, and provides a nurse-friendly chat interface that explains the reasoning behind every suggestion. For example, if a patient presents with SpO2 =88% and chest pain, the system highlights the dangerously low oxygen saturation, assigns an urgent triage score, and suggests immediate interventions aligned with clinical guidelines. In addition, ERgent integrates a readmission score that estimates the likelihood of a patient returning to the hospital based on their clinical data. This predictive insight enables doctors to allocate resources more effectively, providing extra support for high-risk patients and helping to reduce preventable readmissions.

How we built it

We built ERgent on a modular, cloud-native architecture. The frontend was developed using React and TypeScript to deliver a responsive and intuitive user experience tailored to nursing workflows. The backend was implemented with FastAPI and Uvicorn to support high-performance, asynchronous APIs capable of handling concurrent requests. We integrated multiple Azure services, including Azure OpenAI and Semantic Kernel for orchestrating specialized agents, Azure Storage and Blob for secure management of patient data, Azure Identity for authentication and role-based access control, and OpenTelemetry for observability and debugging. A custom database access layer was designed to optimize retrieval of patient histories and provide seamless integration with the patient data agent.

Challenges we ran into

One of our biggest challenges was getting retrieval-augmented generation (RAG) to work reliably. Ensuring that the system retrieved relevant patient histories while avoiding hallucinations required careful prompt design and iterative testing. Setting up the Azure-based agents and orchestrator also required substantial effort, as coordinating multiple specialized agents to work together seamlessly was not straightforward. Finally, integrating the database with the patient history agent was challenging because we had to ensure that retrievals were both consistent and fast, regardless of how large or complex the patient histories became.

Accomplishments that we're proud of

We are proud to have built a functioning multi-agent orchestration system that mirrors real emergency department workflows. One of our greatest achievements was creating a transparent, nurse-friendly interface that does not just provide recommendations but also explains the rationale behind them, fostering trust and usability. We are also proud that ERgent demonstrates the potential for advanced AI systems to improve triage efficiency and patient safety in one of the most critical areas of healthcare.

What we learned

Through this project, we learned how to design and integrate multi-agent systems that can handle complex real-world scenarios. We gained experience with retrieval-augmented generation pipelines and discovered the unique challenges of applying them in sensitive medical contexts. We learned to configure and monitor Azure cloud resources effectively, balancing performance, cost, and compliance. Most importantly, we recognized the importance of human-centered design and transparency in healthcare applications, ensuring that clinicians can both understand and trust AI-driven decision support.

What's next for ERgent

Looking ahead, we plan to extend ERgent with more specialized agents, such as pharmacy or cardiology, to cover broader aspects of emergency care. We also want to expand interoperability with electronic health records (EHRs) to integrate more seamlessly with hospital systems. Usability testing with nursing staff will be a crucial next step, allowing us to refine workflows and optimize the user experience. Finally, we aim to explore compliance and regulatory pathways so that ERgent can move from a prototype into a tool that supports real-world clinical deployment.