Bournemouth, UK · Production engineering since 2012
AI Infrastructure Engineer
Keeping production systems up since 2012, now bringing that experience to AI infrastructure. I work where models meet production: inference performance, Kubernetes and reliable AI platforms.
Production engineering in a regulated environment, with a focus on the point where model performance meets real traffic. These examples describe my contribution, not the work of an entire team.
The problem. Understand how model endpoints behave under sustained concurrent streaming traffic, beyond single-request benchmarks.
My contribution. Designed, built, deployed and operated a distributed Locust load generator on ECS, driving approximately 10,000 simulated concurrent users against models served on A100/H100 hardware.
The output. Analysis of time to first token, time to first 100 tokens, token throughput and streaming stability, used to inform capacity and sizing decisions.
I owned the harness, methodology and analysis. A partner team deployed and operated the model-serving stack.
Operate and support services on EKS, SageMaker, EMR and Databricks. My work covers SLOs, error budgets, incident response and capacity planning, including the on-call rotation that carries those commitments.
Integrated MCP servers and multi-tool orchestration into an agentic incident-triage platform, with attention to the permission model governing access to production systems.
I write about this work as I go: the pieces below cover agentic architecture, RAG economics and running MCP servers in production.
AI/ML Site Reliability Engineer
DevOps Lead
Technology Analyst
Biju Patnaik University of Technology · Odisha, India
Articles on architecture, inference economics and operations are available on dev.to and Medium below.
For conversations about inference performance, AI platform engineering or production reliability, send a note or connect on LinkedIn.