AI, Data & Research - AI & Machine Learning
MLOps Engineer
IC / Practitioner
Independently auditedBuilds systems that learn, and keeps them honest in production. Owns a defined area of the function and the outcomes it produces.
Audited package
Built, independently audited role package. Everything on this page is written for this seat.
Your Stack (Swappable)
This role works against whatever tools you already run. We list the tool categories the role operates in, never a named vendor, so you are never asked to replace a system that works.
- Data Governance & Catalog
- CI/CD
- AI / LLM Platform
- Cloud Infrastructure & IaC
- Feature store
- Observability & Monitoring
- Data Warehouse & Lakehouse
Deliverables
- Evaluation harnesses and results
- Model documentation
- Deployment and monitoring plans
- Capability assessments
Responsibilities
- Build and operate the model CI/CD pipeline - packaging, testing, the automated evaluation and promotion gates - that takes a registered model from a training run to production without hand-deployment
- Own the model registry as the source of truth: every production model versioned and traceable to its training run, data version, code, and environment, with a governed promotion/rollback workflow
- Build and run the serving and inference infrastructure (online, batch, and streaming), including autoscaling, deployment strategies (canary, shadow, blue-green), and one-step rollback, to a defined availability and latency SLO
- Operate the feature store: the online/offline parity contract that computes a feature identically at training and serving time, so train/serve skew is prevented at the platform layer rather than debugged per model
- Build the model-monitoring and observability system that emits drift (data, concept, prediction), data-quality, and production-performance signals with thresholds and alerting, so a decaying model is detected, not discovered
- Guarantee reproducibility: containerized, dependency-pinned training and serving environments and artifact versioning, so any production model can be rebuilt and any result bisected
- Deliver the paved road and the drift signal to the Machine Learning Engineer, AI/ML Engineer, and Data Scientist, and consume the warehouse and feature data the Data Engineer lands - without rebuilding ingestion or touching model accuracy
- Set and defend platform SLOs (serving availability, p99 inference latency, deploy-pipeline reliability, serving cost), investigate serving and pipeline incidents to root cause, and remove the class of failure rather than firefighting each one
The essence of this seat
The MLOps Engineer builds and operates the ML platform - the paved road that every model in the company rides on from a training run to reliable production serving.
Signature move
Reach for the paved road before the one-off. The first question on any ship request is whether it can go through the self-service promotion gate rather than through your hands - a model that routes through you is a bottleneck, so the reflex is to build or use the gate, not to hand-deploy this one and rebuild the stack for the next
What good looks like
replaces the manual deploy with a self-service pipeline (package -> automated eval-gate promotion -> registry -> one-step rollback), not a faster manual deploy; gates on the model's own evaluation and lineage, not a unit test; owns that the gate exists and runs while leaving the acceptance threshold to the modeler
Decision rights
- - the model CI/CD pipeline and its promotion/rollback workflow, the registry design and the gate mechanics, the serving-infrastructure architecture (inference runtime, autoscaling, deployment strategy), the feature-store platform and its online/offline parity contract, the model-monitoring and observability system (which signals, thresholds, alerting), the reproducibility standard (containers, pinned environments, artifact versioning), and the platform SLOs
Working vocabulary
What this role does
The MLOps Engineer builds and operates the ML platform - the paved road that every model in the company rides on from a training run to reliable production serving. They own the model CI/CD that ships a model, the registry that makes every production model traceable, the feature-store operations that keep features consistent between training and serving, the deployment and serving infrastructure that answers inference under load, the monitoring system that watches models for drift and decay, and the reproducibility that lets any production model be rebuilt. This is an individual-contributor role, reporting into an ML Platform Manager or Data Manager, and accountable for platform reliability and the health of the deployment, serving, and monitoring systems - not for a model's accuracy or architecture (that is the Machine Learning Engineer, role 011), not for the ingestion and analytics-warehouse pipelines that feed the data platform (that is the Data Engineer, role 004), and not for shipping the product feature that consumes a model (that is the AI/ML Engineer, role 010). The line that defines this seat: when a model is inaccurate, that belongs to the modeler; when the serving platform is down, the deploy pipeline is broken, or a drifting model slips by unwatched, that belongs to you. What separates great from average is making the second-through-fiftieth model as cheap and safe to ship as the first: a self-service paved road instead of a ticket queue, a promotion gate that runs the model's own evaluation automatically before anything goes live, a rollback that is a button and not an incident, online and offline features computed through one path so skew cannot creep in, and a monitoring system that emits the drift and data-quality signals the moment a model starts to rot. The deliverable is never "the model is deployed" - it is a platform other engineers trust enough to ship on without asking you first.
Best-fit customer
teams putting models into production and needing evaluation discipline
Need more than one role?
Buying the whole team or the complete department lowers your effective cost per role at every tier of the ladder.