Workflow Performance at Scale: Diagnosing and Fixing Latency in Production

Most latency problems don’t show up in dashboards. They hide in timing gaps between steps, redundant data fetches, and cold starts quietly degrading performance until someone finally instruments the right thing.

This session is the engineering story of how DevRev’s team is systematically hunting down latency in production workflow execution without ever touching customer data. From a mysterious ~1 second gap between workflow steps to an O(N·K) data fetch collapsing to O(1), we’ll walk through exactly what the traces revealed and the specific fixes moving p90 and p99 latency.

Watch Ahmed Bashir (CTO at DevRev) and Navneel Mandal (Member of Technical Staff at DevRev) as they show how platform reliability actually improves through observability, smart caching, Lambda provisioned concurrency, and better retry logic for LLM-calling nodes.