We worked with a leading children’s hospital to improve how its intensive care teams monitored critically ill patients.
In the PICU, doctors and nurses were working with data from multiple bedside devices, including patient monitors, ventilators, and infusion pumps. Because these systems operated separately, staff often had to check different screens and manually record readings. Frequent alarms also made it harder to distinguish routine alerts from signs of real deterioration.
We set out to bring these signals into one place, make them easier to act on, and connect the system with the hospital’s existing EMR environment.
The engineering challenge
The project came with a few requirements that made it more demanding than a typical monitoring platform:
- High-volume data: The system needed to handle around 8,000 continuous measurements and up to 100,000 data points per second during peak periods.
- Low latency: Vital signs had to reach a central dashboard in under 2 seconds, across up to 500 beds, while maintaining 99.99% uptime.
- Interoperability: We needed to work with medical devices from different vendors and support data exchange with HIS, EMR, LIS, and PACS through HL7 v2 and HL7 FHIR R4.
- Data privacy: Patient data had to remain on-premise and within the required jurisdiction to meet local data protection requirements, including Decree 13/2023/ND-CP.
Our solution – an event-driven smart ICU platform
We built a centralized Smart ICU monitoring and Clinical Decision Support System (CDSS) using an event-driven microservices architecture.
Technical architecture
- Data ingestion: An IoT Gateway collects vital signs from different medical devices and streams the data through RabbitMQ for reliable, high-speed message processing.
- Real-time monitoring: SignalR keeps the ReactJS dashboard updated with live patient data. We built the dashboard as part of our a work and developed Flutter-based mobile apps for tablets used during bedside rounds.
- Data storage: TimescaleDB stores high-volume time-series data and organizes it into a continuous timeline for each patient, supported by our data engineering expertise.
- Clinical decision support: The CDSS calculates Early Warning Scores (EWS) in real time. A configurable rule engine helps filter routine alerts and escalate signals that may require clinical attention.
An AI-first way of building the product
One part of the project was different from a traditional healthcare development process – how we built it.
Instead of putting a large engineering team behind the project, we had a lean team of 8 Business Analysts and Quality Control specialists take ownership of individual modules.
Claude worked alongside the team as an AI co-pilot. Each member could take a module from functional requirements and prototypes through test cases and production-ready code, with the team reviewing and validating the output at each stage.
This approach helped us move quickly without treating AI as a replacement for domain knowledge or quality control.
The remarkable results
The approach allowed us to move from concept to a working clinical environment in a short time:
- 3-week POC: We delivered a functional proof of concept and deployed it directly in the hospital’s clinical environment within three weeks.
- Clinical integration: The solution was integrated into the Cardiovascular ICU as part of the hospital’s wider EMR ecosystem and later expanded to additional intensive care units.
- Earlier risk detection: Automated risk scoring gave clinicians another way to identify potential deterioration while helping reduce unnecessary alarm noise.
- A repeatable delivery model: We turned the AI-assisted workflow into an internal delivery playbook that can now be applied to other healthcare engineering projects.
The road ahead
The initial implementation gave us a solid foundation for the next stage. We’re now moving toward go-live with the EMR in the Cardiovascular Surgery ICU, with plans to expand the solution to other intensive care units. The next phase will also bring deeper Smart ICU integration into the EMR, stronger AI-powered early warnings for clinical deterioration, and broader adoption of our AI-First delivery approach across EMR and other healthcare projects.
If you’re looking to modernize healthcare workflows with AI and connected systems, we can help you turn the idea into a production-ready solution.
