MLOps Architect

Overview We are looking for a MLOps Architect to own the end-to-end design, evolution, and governance of our Kubeflow-based MLOps platform on Azure. This role is responsible for platform architecture, long-term technical strategy, system reliability, upgrades, cost optimization, and introducing new MLOps capabilities while ensuring zero or minimal impact to running workflows. The role partners closely with ML Engineers, Data Science leaders, Cloud Infrastructure, Software Engineering, and Business stakeholders to translate ML requirements into scalable, secure, and cost-effective platform solutions. Responsibilities Platform Architecture & Strategy Define and own the end-to-end architecture of the Kubeflow-based MLOps platform Establish platform standards for scalability, security, reliability, and multi-tenancy Drive the roadmap for new MLOps capabilities (feature stores, registries, serving, governance) Kubernetes & Kubeflow Lifecycle Management Plan and execute Kubeflow and Kubernetes, right-sizing) Cross-Functional Leadership Act as technical advisor to Data Science, MLE, Infrastructure, and Software teams Collaborate with Business stakeholders to align platform capabilities with business goals Mentor junior MLOps engineers and set engineering best practices Qualifications 7+ years of experience in MLOps, Platform Engineering, DevOps, or ML Infrastructure Deep expertise in Kubeflow internals, Kubernetes architecture, and multi-tenant ML platforms Strong architectural experience on Azure (AKS, Networking, Security, Governance, Cost Management) Proven experience leading platform upgrades (Kubeflow & Kubernetes) with minimal disruption Advanced CI/CD, GitOps, and Infrastructure-as-Code experience (Terraform, Helm, Argo CD) Experience designing platform-level observability, governance, and FinOps frameworks Strong leadership, technical decision-making, and stakeholder management skills What makes us different? * Hybrid work model: combination of remote and collaborative

office experience to enable innovation * Entrepreneurial environment in leading international company * Professional growth possibilities & learning opportunities * Variety of benefits to support your physical, emotional and financial wellbeing * Volunteering opportunities to help external communities About PepsiCo We believe that culture should be at the cornerstone of everything we do at PepsiCo. We are agile, innovative and not afraid of failure. We want our team to come to work every day excited to explore new ways to bring enjoyment, refreshment and fun to the world. PepsiCo Positive (pep+) is the future of our organization - a strategic end-to-end transformation, with sustainability at the center of how we will create growth and value by operating within planetary boundaries and inspiring positive change for the planet and people. So, if you’re ready to be a part of a playground for those who think big, we’d love to chat. *We encourage the diversity of applicants across gender, age, ethnicity, nationality, sexual orientation, social background, religion or belief and disability #LI-Hybrid

version upgrades with minimal workflow disruption Evaluate and introduce new Kubeflow components and ecosystem tools Ensure backward compatibility and migration strategies for existing workflows CI/CD, GitOps & Automation Design advanced CI/CD and GitOps workflows for platform and ML assets Standardize environment promotion (dev * staging * prod) for ML workflows Define best practices for infrastructure-as-code and platform automation Observability, Security & Governance Architect platform-wide observability (metrics, logs, traces, ML-specific monitoring) Implement governance controls for access, data usage, and environment isolation Ensure compliance with enterprise security standards and cloud governance policies FinOps & Cost Attribution Design and operate cost attribution models per workflow, team, or business unit Partner with Finance and Business teams to provide cost visibility and optimization insights Drive cost optimization strategies (autoscaling, spot instances