Dinesh kumar.N

On Musecool, a music AI product, I led the migration of a monolithic three-tier application into four containerized microservices, improving scalability and fault isolation. I implemented GPU auto-scaling with KServe and Knative using a scale-to-zero strategy, which cut infrastructure costs by around 80% while keeping on-demand inference fast. I also integrated MLflow for experiment tracking and model versioning, used Optuna to keep model hyperparameters tuned as input data evolved, and designed a Redis and RabbitMQ buffering layer to deliver music recordings to inference services smoothly without overload. Along the way, I built CI/CD pipelines that eliminated 90% of manual deployments and planned a near-zero-downtime migration from GCP to AWS using a Blue/Green strategy. On Dataflow, I architected a production-grade microservices platform running across dev, staging, and production Kubernetes clusters, supporting 9+ highly available deployments and 100+ concurrent users. I reduced pod startup latency by 75% through container image snapshotting and warm-up instances, and cut pod initialization time by 60% by replacing NFS-based Python environment mounts with a SquashFS distribution system using Kubernetes mount propagation. I enforced Infrastructure as Code with Terraform and Helm and engineered an automated billing pipeline that aggregates data from six platform features. My core toolkit includes Kubernetes, Helm, KServe, Istio, MLflow, Terraform, AWS, GCP, Python, Bash, Git, and Linux.