MLOps: market

Vacancy

an open job that cleared the quality gate and lists at least two technologies

Observed candidate

a candidate whose stack contains this technology; an aggregate — not a single record leaves the perimeter

Matchable candidate

not counted yet

Window

jobs open at the moment the snapshot was built

Junior MLOps — typical entry: 1) DevOps / SRE Middle + ML interest, 2) ML Engineer Middle + infra interest, 3) Backend Senior + ML knowledge (via courses). Junior → Middle jump — after the first end-to-end ML platform deployment (Feature Store + Model Registry + Serving stack + Monitoring). Middle → Senior — multi-team platform ownership + cost optimisation (typical mandate: 30-50% reduction in compute costs via quantisation + caching + spot instances). Senior → Staff / Principal — ML platform architecture for multi-product organisation + integration strategy with DevOps / SRE / Data Platform teams. Career flow: DevOps Middle / ML Engineer Middle → Junior MLOps (1-2 years) → Middle (2-3 years) → Senior → either Staff / Principal MLOps (deep), MLOps Tech Lead (management), or Founder at an ML-platform startup.

Top 5: mlops, ci/cd, aws, cloud, python. Python (primary — for custom infrastructure), Go bonus (for performance-critical infra components — e.g. serving proxies). Docker mastery (multi-stage builds for optimised ML images, GPU containers). Kubernetes mastery (production standard for ML serving — Deployment / StatefulSet / DaemonSet / Job / CronJob, RBAC, Network Policies, Resource Quotas, HPA / VPA / Cluster Autoscaler). GPU operator (NVIDIA GPU Operator for K8s GPU scheduling). K8s operators for ML: Kubeflow Training Operator + Spark Operator + Argo Workflows. Helm + Kustomize + ArgoCD / FluxCD — GitOps standard. Terraform / Pulumi / CloudFormation (IaC for cloud-managed ML — AWS / GCP / Azure resources). MLflow mastery (open-source standard for model registry + tracking — Senior owns MLflow Plugins customisation). Kubeflow + Kubeflow Pipelines (K8s-native ML orchestration — large enterprise often requires). Workflow orchestrators: Airflow + Prefect + Dagster (one deeply — Dagster rising 2026). Feature stores: Feast (open-source) or Tecton (managed) — must for Senior MLOps. Model serving: Triton Inference Server (NVIDIA — multi-framework + dynamic batching) + BentoML (Python-friendly) + TorchServe + KServe (K8s-native serving). LLM-specific serving: vLLM (PagedAttention — record for high-throughput LLM serving), Text Generation Inference (TGI) from Hugging Face, SGLang (programming model for LLM serving + caching), Ollama (local LLM serving). Observability: Prometheus + Grafana + Loki + OpenTelemetry. ML monitoring + drift detection: Evidently AI + WhyLabs + Arize AI + Fiddler (one deeply). Data version control: DVC + LakeFS. Cloud-managed ML: one of AWS SageMaker / GCP Vertex AI / Azure ML / Databricks deeply. Distributed compute platforms: Spark + Ray mastery. Cost optimisation: spot instances strategy, GPU sharing (NVIDIA MIG), quantisation integration, caching strategies (Redis for inference results), batch vs real-time inference cost analysis.

DevOps Engineer — focus on general application infrastructure (CI/CD + monitoring + cloud + Kubernetes + IaC). Stack: Docker / K8s / Terraform / Prometheus / Grafana / Ansible / Jenkins / GitHub Actions / cloud (AWS / GCP / Azure). See DevOps Engineer (general) (when the page ships). Platform Engineer — DevOps + internal-tooling focus. Builds a developer platform for engineers (K8s clusters management + ArgoCD + IDP — Internal Developer Platform — Backstage / Port / Cortex). MLOps Engineer (this page) — DevOps + ML-specific focus. Builds an ML platform for DS / ML Engineer / Research teams. On top of the DevOps stack: MLflow / Kubeflow / Feature Stores / Model Serving (Triton / vLLM) / ML Monitoring (Evidently / WhyLabs) / cloud-managed ML (SageMaker / Vertex AI). ML Platform Engineer — Senior MLOps with focus on reusable platform-as-product (for the whole organisation, not a single team). Career overlap: DevOps Middle → MLOps Junior — 6-12 months (learn ML stack + Kubeflow / MLflow + cloud-managed ML). MLOps Senior ↔ Platform Engineer Senior — easy lateral move (much shared stack). ML Engineer Middle → MLOps Junior — 4-8 months (already knows ML stack, need to deepen Kubernetes + Terraform + observability). DevOps → MLOps — more distant than Platform Engineer pivot (ML knowledge gap).

Decision tree for ML platform choice 2026: 1) MLflow (open-source, vendor-neutral) — best for experiment tracking + model registry + simple deployment. Lightweight. Self-hosted or managed (Databricks). Use case: foundation of ML platform for most organisations. Default 2026 for team-level ML tracking. 2) Kubeflow + Kubeflow Pipelines (open-source, Kubernetes-native) — best for enterprise K8s-first organisations that want a full ML-platform stack without vendor lock. Heavier setup. Includes: Pipelines (workflow) + Training Operator (distributed training) + Serving (KFServing → KServe) + Notebooks. Use case: large enterprise + Kubernetes-mature team + multi-cloud / on-prem requirement. 3) AWS SageMaker (managed) — best for AWS-native organisations. Comprehensive: SageMaker Studio (IDE) + Training Jobs + Hyperparameter Tuning + Pipelines + Endpoints (real-time + batch) + Model Registry + Feature Store + Clarify (bias detection). Vendor lock to AWS. Use case: AWS-first organisation + small ML team — pays off thanks to feature count in one platform. Cost-heavy at scale (expensive). 4) GCP Vertex AI (managed) — Google's SageMaker equivalent. Strong on: Vertex AI Workbench (Jupyter cloud), Pipelines (Kubeflow-based — portable), AutoML, TPU access (for research / large training), Model Garden (pre-trained models). Use case: GCP-first organisation + Kubeflow portability matters. 5) Azure ML (managed) — Azure equivalent. Strong on: Designer (visual ML pipeline builder), AutoML, MLOps capabilities, Azure-native integration. Use case: Azure / Microsoft-first organisation. 6) Databricks (managed multi-cloud) — best for Spark-first ML workloads + Lakehouse. MLflow creators. Premium pricing. Use case: data-heavy organisations with Spark / Lakehouse pattern. 7) Specialised platforms: Weights & Biases (best UX for experiment tracking — research favourite), Comet (W&B alternative), Neptune, Determined AI (distributed training platform), Run:AI (GPU orchestration — acquired by NVIDIA), Truefoundry + Bento (modern model serving). Default 2026 for new ML team: MLflow self-hosted (or Databricks managed if Spark-heavy) + Kubernetes serving (Triton / vLLM) + Feast (feature store) + Prometheus + Grafana + Evidently AI. Default for AWS-native enterprise: SageMaker (if cost-acceptable + small team) or MLflow on EKS + custom services (cheaper at scale). Default for Kubernetes-mature enterprise: Kubeflow + MLflow.

Yes, 35% of MLOps jobs are full-remote or hybrid. MLOps work is cloud-based (all infrastructure in cloud K8s / managed services). Outsourcing shops (EPAM AI Platform / Luxoft / Andersen) — almost always remote on US AI projects. Russian product companies — hybrid or remote after probation. Russian banks (Sber AI banking / VTB) — hybrid/office due to security compliance. International ML-platform companies (Truefoundry / Bento / Weights & Biases / Comet / Domino Data Lab / Determined AI / Run:AI) — full-remote standard. Big Tech ML Platform — hybrid-standard, partly remote. Relocant hubs for MLOps: Poland (Warsaw / Krakow — AI Platform-friendly), Germany (Berlin / Munich), Canada (Toronto), Serbia, Georgia, UAE. English for international MLOps remote — must (premium +30-50%, and MLOps involves intensive cross-team collaboration in English).

MLOps Engineer (general) — can work in any ML team (product team / research lab / AI consultancy). Stack: universal MLOps. ML Platform Engineer — Senior MLOps with focus on reusable platform-as-product for the whole organisation (not for a single team). On top of that: internal API design (for other ML Engineers / DS), self-service tools (templates / scaffolding / CLI tools — like generate-new-model-project), SLA management for platform components (uptime / latency / throughput targets), cost-chargeback systems (per-team compute usage tracking + billing), onboarding documentation + training programmes for other ML team members, backwards-compatibility management for shared platform components, open-source contribution to platform components (e.g. Feast / Kubeflow contributions). ML Platform Engineer salary is 15-25% above Senior MLOps. Staff / Principal MLOps — even higher, typically managing an ML platform team (3-7 MLOps engineers + integrations with DevOps / SRE / Data Platform leads). Career choice: MLOps Engineer if you like being hands-on with a new problem every day, ML Platform Engineer if you like building reusable systems and working with multiple consumer teams. All three are legitimate Senior+ paths.

Roadmap: 1) Strong DevOps base — Linux + networking + bash mastery, Docker mastery (multi-stage builds + layer optimisation + security scanning), Kubernetes mastery deep (Deployment / Service / Ingress / PVC / RBAC / Network Policies / HPA + custom CRDs basics). Books: "Kubernetes in Action" Lukša, "The Kubernetes Book" Poulton. 2) IaC mastery: Terraform deep (modules / state management / workspaces) + understanding cloud-provider-specific patterns. HashiCorp Learn course (free). GitOps patterns with ArgoCD / FluxCD. 4) Cloud platform deep — pick one of AWS / GCP / Azure and pass Solutions Architect Associate / Professional cert. 5) ML basics — minimum: Python + pandas + scikit-learn + PyTorch (basics) + understand training / inference loops. fast.ai "Practical Deep Learning" Part 1 course. "Designing Machine Learning Systems" Chip Huyen — MLOps must-read 2026 (for understanding the ML workflow from training to deployment). 6) Experiment tracking + model registry: MLflow mastery deep (Tracking + Registry + Projects + Models). Set up self-hosted MLflow on EKS / GKE / AKS. 7) ML serving stack: Triton Inference Server (multi-framework + dynamic batching) + BentoML + KServe. LLM serving: vLLM mastery (PagedAttention + continuous batching) + TGI + SGLang. 8) Feature Store: Feast (open-source) — set up end-to-end on a test project (offline + online + retrieval). 9) Orchestration: one of Airflow / Prefect / Dagster + Kubeflow Pipelines basics. 10) ML monitoring: Evidently AI (drift detection + data quality) + WhyLabs or Arize AI or Fiddler — one deeply. 11) Pet project: full ML platform deployment (K8s cluster + MLflow + Feast + Triton + Prometheus + Grafana + Evidently AI) with simple model lifecycle (train → register → deploy → monitor → retrain). Document in GitHub README. 12) Open-source contribution — Feast / Kubeflow / MLflow / vLLM / Triton — sharply increases MLOps resume value. Russian courses: Karpov.Courses "MLOps", Otus "MLOps", SkillFactory MLOps track, Slurm MLOps intensive. International (EN): "Made with ML" MLOps course Goku Mohandas (free, comprehensive), Hugging Face MLOps course, Coursera "MLOps Specialization" DeepLearning.AI, Databricks Academy MLOps. Must-read books: "Designing Machine Learning Systems" Chip Huyen, "Machine Learning Design Patterns" Lakshmanan, "Reliable Machine Learning" Cathy Chen. DevOps Middle / ML Engineer Middle → Junior MLOps — 4-10 months. Backend Senior → Junior MLOps — 8-14 months (need to learn ML stack).

188 active open MLOps positions — growing ML-infrastructure niche. Geography: Russia / Poland / remote. The real market is broader thanks to the international remote segment (ML-platform companies — Truefoundry / Bento / W&B / Comet / Determined AI / Domino Data Lab + AI-startup MLOps teams — full-remote standard). Time to close a Senior MLOps role — 5-10 weeks (faster than Research, slower than DevOps general due to rare ML-skill combination). MLOps — the most "recession-resistant" AI/ML niche (companies cut research first, MLOps continuously needed for production).

A Senior MLOps owns the full ML-platform engineering cycle + technical leadership. Programming mastery: Python deep + Go bonus for performance-critical infra. Docker mastery: multi-stage builds, layer optimisation, GPU containers, security scanning (Trivy / Snyk). Kubernetes mastery deep: Deployment / Service / Ingress / PVC / RBAC / Network Policies / HPA / VPA / Cluster Autoscaler, custom CRDs basics, GPU Operator + Kubeflow Training Operator + Spark Operator + Argo Workflows. IaC mastery: Terraform deep (modules / state / workspaces), one of Pulumi / CloudFormation bonus. GitOps mastery: ArgoCD / FluxCD + Kustomize / Helm patterns. Cloud platform deep: one of AWS / GCP / Azure at Solutions Architect Professional level. MLflow mastery: deep customisation + Plugins + multi-tenancy patterns. Kubeflow Pipelines: enterprise-scale deployment + custom components + multi-tenancy. Feature Store mastery: Feast deep (offline + online + retrieval + monitoring + multi-tenancy) or Tecton (managed). Model serving mastery: Triton Inference Server deep (ensembles + dynamic batching + model repositories) + BentoML + KServe. LLM serving mastery (2026 must): vLLM deep (PagedAttention + continuous batching + speculative decoding + KV cache optimisation) + TGI + SGLang + Ollama. Distributed compute platforms: Spark + Ray mastery (Spark Operator on K8s, Ray Train + Ray Serve). Observability deep: Prometheus + Grafana + Loki + OpenTelemetry + custom metrics + alerting strategies. ML monitoring: Evidently AI + WhyLabs / Arize AI / Fiddler integration patterns. Cost optimisation mastery: spot instances strategy + GPU sharing (NVIDIA MIG) + quantisation integration (INT8 / FP16 / FP4) + caching strategies (Redis for inference results) + batch vs real-time inference cost analysis + reserved instance planning. Typical Senior MLOps mandate: 30-50% reduction in compute costs through a combination of these techniques. Security mastery: secrets management (Vault / cloud KMS), network policies, supply chain security (signed images), data privacy (PII handling in ML pipelines). System design for ML platforms: design ML platforms on a whiteboard at scale 100M+ predictions/day + multi-tenant + multi-region. Soft: ADRs writing for platform decisions, technical writing (platform documentation + onboarding materials), code review for multiple teams, mentoring Middle MLOps, cross-team collaboration (DevOps / SRE / Data Platform / ML Engineer / DS / Research teams). English for Senior+ MUST — MLOps involves intensive cross-team work in English. Optional bonus: open-source contributions to Feast / Kubeflow / MLflow / vLLM / Triton — sharply increase market value for frontier-AI labs + ML-platform companies.