Senior AI/ML Engineer

Architecting and deploying ML solutions using SageMaker Pipelines for automated workflows, configuring EC2/EKS for model serving, managing IAM roles for secure access, and setting up VPCs for network isolation. You're comfortable optimising instance types (e.g., Graviton vs. Intel, Spot vs. On-Demand) for cost and performance.

Designing novel model architectures, implementing complex algorithms, optimising model performance with tools like PyTorch Lightning, and contributing to shared internal ML libraries. You're also a wizard with data manipulation and feature engineering using pandas and NumPy.

Designing and implementing end-to-end experiment tracking, model registry, and deployment workflows. You'll be integrating these MLOps tools into our CI/CD pipeline to ensure reproducibility and automation.

Writing efficient, multi-stage Dockerfiles for ML environments. Deploying and debugging containerised applications and models on a Kubernetes cluster using `kubectl` and Helm charts. You understand how to manage pods, services, and deployments for ML workloads.

Designing and optimising complex ETL/ELT pipelines for ML data. This includes writing advanced Spark/SQL queries, managing cluster configurations, and scheduling jobs for efficiency and cost within our data platform.

Managing complex merges, resolving conflicts, enforcing branching strategies (like GitFlow), and performing thorough, constructive code reviews for your team. You're a master of version control.

Breaking down epics into detailed stories, managing sprint planning for your workstreams, and writing comprehensive technical design documents. You'll be using these to track progress and share knowledge.

• Product Management (to understand what customers really need)
• Software Engineering (to integrate your models into our products)
• Data Engineering (to get the clean, reliable data you need)
• UX/UI Design (to ensure the AI experience is intuitive)
• Director of AI/ML (for strategic alignment and technical direction)

• Key vendors for specific ML tools or cloud services (occasionally)
• Academic partners (if we're collaborating on research)
• Senior clients (when you need to explain complex model behaviour in simple terms)

As our AI footprint grows, the complexity of managing models in production skyrockets. You'll need to master advanced techniques for continuous integration/continuous deployment (CI/CD) specifically for ML, robust monitoring, auto-scaling inference services, and disaster recovery strategies. It's about making our AI systems truly enterprise-grade.

GitOps for ML · Canary & Blue/Green Deployments for Models · Model Observability & Alerting · Distributed Training Optimisation

• This week: Deep-dive into our current CI/CD pipelines for ML. Identify bottlenecks or manual steps.
• This month: Propose and implement one significant improvement to our MLOps workflow (e.g., automated model validation in CI).
• Month 2: Research and prototype a new deployment strategy (e.g., canary releases) for one of our less critical models.
• Month 3: Lead a knowledge-sharing session on advanced MLOps topics for the team.

Quick win: Automate a manual step in your current model deployment process. Even a small script can save hours and reduce errors.

With increasing regulatory scrutiny and the sensitive nature of data used in AI, understanding how to build secure and privacy-preserving ML systems is paramount. This includes protecting models from adversarial attacks, ensuring data anonymisation, and implementing robust access controls.

Adversarial Machine Learning · Differential Privacy · Federated Learning · Secure Multi-Party Computation (SMC)

• This week: Read up on common adversarial attacks (e.g., FGSM) and basic defences.
• This month: Review our current data anonymisation practices for ML datasets. Identify potential gaps.
• Month 2: Participate in a security review of one of our production ML services. Learn from the security team.
• Month 3: Research and present a brief on the implications of a specific privacy-preserving ML technique for our business.

Quick win: Ensure all data used in your models is appropriately anonymised and access-controlled. It's a fundamental step.