I work on large-scale distributed systems that run continuously across regions and operate under real production constraints. My focus is on the parts of systems that only become visible under stress: latency spikes, partial failures, backpressure, and the operational edges that determine whether a system is resilient or fragile in practice.
My work spans data streaming, analytics infrastructure, and distributed systems that support both batch and real-time machine learning workloads. I’m particularly interested in how data moves through complex systems—and what it takes to keep those systems correct, observable, and predictable as they scale.
At Confluent, I focus on real production use cases across Apache Kafka, Apache Flink, and emerging agentic patterns. This work sits at the intersection of system design and operational reality: how AI state is modeled, how failures are surfaced and reasoned about, and how to design pipelines that continue to behave predictably under imperfect conditions.
A consistent theme across my work is translating between architectural intent and what actually happens in production. I spend a lot of time with teams building large data and AI platforms, helping reduce accidental complexity so systems remain understandable and operable as they grow.
More recently, I’ve been focused on real-time data systems that underpin ML and retrieval workloads. The interesting problems are the guarantees required from the ML pipeline & data layer: freshness, consistency, recoverability, and the ability to reconstruct state when systems drift. This often leads to work on self-healing pipelines, observability as a core design primitive, and tighter feedback loops between production behavior and system design.
I’m the author of High Performance Spark (2nd Edition) and Scaling Machine Learning with Spark, both grounded in building and operating distributed systems in real production environments.