Senior AI Data Scientist

You'll be building complex models from scratch, optimising performance, and diving deep into the internals of one or more deep learning frameworks. You'll be expected to contribute to internal libraries and set coding standards.

You'll be designing and implementing end-to-end training and deployment pipelines, managing compute resources (e.g., EC2, A100s) effectively, and troubleshooting production issues. You'll know the ins and outs of these platforms.

You'll be writing and optimising complex PySpark jobs for large-scale feature engineering, designing data schemas for ML tables, and querying massive datasets efficiently. You'll understand the performance characteristics of distributed systems.

You'll be designing the experimentation framework, implementing CI/CD pipelines for models using tools like GitHub Actions or Jenkins, and ensuring robust model versioning and monitoring. You'll be a champion for reproducible ML.

You'll be writing production-ready, multi-stage Dockerfiles from scratch, and deploying services using Kubernetes (K8s) manifests. You'll understand how to containerise your models for scalable deployment.

You'll be managing complex merges, resolving conflicts, and using advanced Git commands like `git rebase` effectively. You'll conduct thorough code reviews for peers and help maintain a clean codebase.

While your focus isn't building dashboards all day, you'll be building interactive dashboards to communicate model performance, business impact, and key insights to stakeholders. You'll need to tell a clear story with data.

• Product Leads (to understand business problems and requirements)
• Engineering Managers (for model deployment and infrastructure support)
• Other Senior Data Scientists (for peer review and technical collaboration)
• Business Analysts (to refine data requirements and interpret results)
• Operations Teams (who'll use your models day-to-day)

• Cloud Platform Vendors (e.g., AWS, GCP for technical discussions)
• Specialised Tooling Providers (for MLOps or niche AI solutions)
• Academic Researchers (occasionally, for cutting-edge techniques)

Getting models into production is still the hardest part of data science. As our AI systems become more complex and critical, you'll need to master advanced MLOps patterns, including robust CI/CD for ML, automated model retraining, and sophisticated A/B testing frameworks for models. This is about building truly resilient and scalable AI infrastructure.

Feature Stores · Model Observability & Anomaly Detection · A/B Testing for ML Models · Infrastructure as Code (IaC) for ML · Model Governance & Auditability

• This week: Review our current MLOps pipeline and identify one area for improvement.
• This month: Take a deep-dive into a feature store solution (e.g., Feast, Tecton) and present its pros and cons to the team.
• Month 2: Lead the implementation of an improved model monitoring dashboard for one of your production models.
• Month 3: Work with engineering to define requirements for an automated A/B testing framework for our ML services.

Quick win: Automate one manual step in your current model deployment process using a simple script or GitHub Action.

While not every problem needs deep learning, for those that do, you'll need to move beyond standard architectures. This means understanding how to design custom neural networks, implement advanced optimisation techniques, and effectively use specialised hardware (GPUs, TPUs) to train massive models efficiently. You'll be pushing the boundaries of what our models can achieve.

Custom Layer Design & Activation Functions · Distributed Training Strategies · Model Compression & Quantisation · Self-Supervised Learning & Foundation Models · GPU/TPU Optimisation

• This week: Experiment with a custom loss function or activation function in one of your deep learning models.
• This month: Read a recent paper on distributed deep learning training and try to replicate a small example.
• Month 2: Investigate model compression techniques for one of our existing deep learning models to improve its inference speed.
• Month 3: Propose a project where self-supervised learning could potentially reduce our reliance on large labelled datasets.

Quick win: Run a simple deep learning model on a GPU instance in the cloud and compare its training time to a CPU, understanding the cost-performance trade-offs.