Custom ASICs generated performance data, but no pipeline existed to capture it in real-time. Counter metrics stayed locked in hardware, forcing teams to poll manually or skip monitoring entirely. Machine learning models couldn't train on live switch behavior, and network diagnostics relied on stale logs. 

Solution 

Calsoft built a streaming telemetry system that pulled high-frequency counters from ASICs and normalized them for ML ingestion. 

• Direct ASIC interface for low-latency counter extraction 
• Kafka pipeline for scalable event delivery to analytics systems 
• Spark transformation layer to structure data with timestamps and metadata 
• Schema validation and CSV export for cross-team compatibility 

The system delivered continuous telemetry from edge switches to centralized platforms without manual polling.