Overview
A digital payments startup processed 50 000 TPS globally but flagged too many false positives, slowing investigation teams.
Solution
- Streamed events with Azure Event Hubs + Kafka.
- Trained real‑time anomaly models in Azure Machine Learning with AutoML.
- Surfaced alerts in Power BI Real‑Time dashboards.
- Embedded model‑explainability via SHAP in the investigator portal.
Results
45 %
Faster incident triage
‑32 %
False positives
99.98 %
Uptime SLA
Tech Stack
Event Hubs • Azure ML • AutoML • Azure Functions • PostgreSQL Hyperscale • Power BI