About this role
Job title: Senior Machine Learning Operations Engineer
About the Role Paramount is hiring a Senior Machine Learning Operations Engineer to own the operational layer around our personalization and recommendation ML systems. You will ensure model traceability, robust monitoring, and proactive incident response, sitting with DevOps and collaborating with ML engineers to maintain trusted, scalable production ML pipelines.
What You'll Do
- Own model traceability: ensure clear lineage for every production model (data trained on, code produced, validation passed, and performance) and drive adoption of tooling for versioning, metadata, and model registries.
- Build end-to-end monitoring: monitor data arrival, feature distribution stability, model metrics, and serving latency against SLA; own this end-to-end.
- Partner with Data Engineering on data quality: surface data quality issues, detect drift in upstream sources, and ensure features stay fresh and reliable.
- Detect issues proactively: track drift over weeks, flag slow degradation before it crosses a threshold, and surface feature freshness problems before they cascade.
- Build diagnostic tooling: ensure the right context is logged at each stage (candidates, features, serving context) and build dashboards to tie it together for root cause analysis.
- Own incident response for ML systems: maintain rollback playbooks and predefined hotfix strategies with quantified trade-offs; own automated gates that block bad deployments; run post-mortems and close gaps.
- Coordinate on post-deployment metrics: work with ML engineers, data engineers, and stakeholders to define what metrics to collect after deployment and why they matter.
What We're Looking For
- 5+ years in ML engineering, applied ML, or a related ML role, with demonstrated experience on the operational side of monitoring, reliability, deployment, or incident response.
- Has built or operated model registries, ML monitoring systems, or production ML pipelines.
- Understands ML systems end-to-end — not just the infra layer, but why a stale feature or a shifted distribution matters.