Berkay Şahin · MLOps Engineer
Machine learning models are easy to train and hard to keep running. I build the pipelines that move them into production, the cloud they run on, and the monitoring that pages you before the user does.
End-to-end ML pipelines on Databricks, deployed through GitLab CI/CD with Databricks Asset Bundles, authenticating as service principals over OAuth M2M rather than someone's personal token.
Provisioning and running the cloud resources ML workloads sit on, and keeping the bill from quietly growing past what the workload is worth.
Cluster management for batch and training workloads — scheduling, lifecycle, and the container images they run from.
Centralised logs through Promtail into Grafana Loki, metrics on Prometheus, and dashboards that answer the question you actually have at 3am.
I read everything that arrives and reply to anything that isn't a template.