At carVertical, I architected and implemented a production MLOps platform on Amazon EKS for GPU-based machine-learning workloads. I worked with ML engineers to turn model-development requirements into production infrastructure, including model delivery, GPU-aware scheduling, and deployment safety.
I built inference infrastructure with NVIDIA Triton Inference Server, NVIDIA GPU Operator, and KAI Scheduler, and introduced fractional GPU scheduling. I also implemented Datadog-driven remediation to recycle model pods under GPU memory pressure before out-of-memory failures.
At Cloudvisor, I worked directly with customer CTOs on AWS architecture, proposals, and hands-on delivery. I re-engineered a Ruby on Rails data-processing workload in Python and Pandas, reducing its execution time from more than 24 hours to approximately 3.5 hours.
