ICU Companion

ICU monitors emit vitals every second. Clinicians needed that data live — as smooth graphs, not jumpy plots — plus lab results and early-warning scores, without the pipeline or the UI falling behind.

Vitals arrive every second and cannot be dropped. The UI has to stay smooth under that rate, and historical readings still need a durable trail.

Built a real-time monitoring path: Kafka ingests the stream, consumers process 1-second updates, data is aggregated for 1-minute persistence, and Redis caches large datasets for the API. The React dashboard uses Redux to apply high-frequency updates without unnecessary re-renders, and interpolates between incoming points so graphs stay readable as vitals change rapidly. LIS provider APIs feed laboratory reports into the same workflows. An AI-assisted layer enhances SOFA and APACHE II scoring, sepsis risk prediction, and early-warning alerts for the Central Monitoring System.

High-frequency vital updates would otherwise thrash the React tree. Redux was used to patch monitoring views incrementally, and interpolation between incoming points kept graphs readable instead of stair-stepping.

1-second processing on the hot path, 1-minute aggregation for persistence, and Redis in front of large reads so the database is not queried for every dashboard refresh.

Real-time monitoring is a frontend problem as much as a streaming one — if the UI re-renders on every event, the pipeline's throughput does not matter.