Data Scientist

Job Description

COO Risk sits right at the heart of the Bank’s complexity. We translate our rather complex banking business exposures into impact on risk weighted assets and provisions in line with the constantly changing business desires and the latest  regulatory requirements. We are an integral part of the Credit Risk Modelling organization where we manage our processes from credit risk modeling approach definition all the way up to post production monitoring. We hereby offer a unique opportunity to enter one of Europe’s most appealing Banks at a key position that enables you to contribute, collaborate and grow!

In a nutshell, it is all about collaboration and understanding the end to end complexity of the Bank in particular it’s products and related credit risk exposures [including corresponding processing activities and the way these are underpinned with state of the art technology solutions]. If you have a natural drive to collaborate and a sense of urgency to conclude and deliver we would be very keen to get in touch!            

The majority of our use cases run on the Vista platform (GCP, BigQuery, Vertex AI, Gemini), with Microsoft Copilot Studio, M365 Copilot and Power Platform as a strategic second stack for collaboration-layer agents. We deliver a mix of classical machine learning models, agentic AI systems, and full applications wrapped around both - whichever is the right tool for the problem in front of us.

This is an individual contributor role for a data scientist who can do all three: ship a production ML model, design and operate a multi-agent system, and build the application layer around either.

Who are we looking for?

• 3+ years as a data scientist or ML engineer, with 1+ years building and shipping production GenAI / agentic systems on top of a real classical ML foundation.
• You have personally taken both classical ML models and agentic applications to production with real users, real SLAs, and real failure modes you had to debug.
• You are multi-cloud by instinct. You move comfortably between Google Cloud and Microsoft Azure stacks and have strong opinions on when each fits.
• You think in systems and evals, not demos or notebooks. You can talk about regression suites, calibration, drift monitoring, and where each one breaks.
• You can build the application around the model, not just hand a file to someone else.
• You are comfortable in a regulated environment and understand why auditability, traceability, and human-in-the-loop are first-class design constraints, not afterthoughts.

What You’ll Do

• Build classical ML models that solve real CRO problems: classification, regression, time series, anomaly detection, clustering, NLP. End-to-end ownership from problem framing through deployment and monitoring.
• Architect and ship multi-agent systems on Vista (GCP) as the primary platform: Vertex AI Agent Builder, Gemini, BigQuery as the analytical and grounding layer, Vertex AI Search, Cloud Run, and the surrounding GCP services. This is where the majority of our agents live.
• Build collaboration-layer agents on Microsoft: Copilot Studio, M365 Copilot, Power Platform. Intent routing, tool use, fallback handling, escalation, and orchestration in the daily-work surface our colleagues already use.
• Develop applications and agentic harnesses around both classical models and LLMs: APIs, services, internal tools, and the orchestration logic that turns a model into a product CRO colleagues actually use every day.
• Own the retrieval and grounding layer end-to-end: BigQuery and Vertex AI Search on the GCP side, Azure AI Search on the Microsoft side. Hybrid retrieval, chunking strategy, reranking, citation handling, and grounded-answer evaluation. You will defend your design choices with numbers.
• Build the evaluation backbone that gates every release: golden datasets, regression suites, LLM-as-judge with human calibration, classical ML evaluation rigour (calibration, fairness, robustness), and drift monitoring in production.
• Partner with risk domain experts across credit, market, operational, and compliance risk. You translate their workflows into models, agents, or applications that genuinely shorten cycle time.
• Build the scalable platform that supports phased rollouts, A/B testing, learning loops, compliance-by-design controls, and reuse across the bank. You will set patterns the rest of the function adopts.
• Champion governance and EU AI Act alignment: lifecycle controls, model cards, change management, prompt and model versioning, data lineage, and the auditability story for second and third line.
• Mentor and lift the bar. Code review, design review, and pairing with the broader CRO data science and AI engineering community.

Your profile

Engineering fundamentals

• Python at senior level: async, packaging, testing (pytest), typing, modern tooling. Not scripting-level Python.
• SQL at senior level, ideally on BigQuery.
• Application development: you can build and ship a production service or internal app, not just a notebook. APIs, FastAPI / Flask or equivalent, basic frontend awareness.
• YAML, JSON (adaptive cards and data handling), HTML/CSS, JavaScript/TypeScript.
• Git, CI/CD on GitHub Actions or Cloud Build, code review discipline, infrastructure as code awareness (Terraform a plus).
• APIs and integration patterns, accessibility standards (WCAG), and localisation best practices.

Classical machine learning

• Production ML experience with the standard toolkit: scikit-learn, XGBoost / LightGBM, statsmodels, and modern equivalents. Comfortable with classification, regression, time series, and basic NLP.
• Statistical foundations: hypothesis testing, A/B testing, causal inference basics, calibration, uncertainty quantification.
• Feature engineering and model evaluation as a craft, not a checklist. You know why a model is good or bad, not just its AUC.
• Pandas, NumPy at expert level. You write efficient, readable, testable data code.

GenAI and agentic engineering

• Production-grade prompt design for agents: system prompts, tool/function calling, structured output, prompt versioning, and prompt regression testing.
• Eval and observability for LLM systems: tracing, golden datasets, LLM-as-judge with human calibration, regression gates in CI, and drift monitoring in production.
• Multi-model fluency: Gemini and GPT-class models in production. You know the tradeoffs, the failure modes, and the cost curves.
• Agentic harnesses and orchestration: you have built the application layer around LLMs, not just called the API once.

GCP / Vista platform

• Vertex AI at expert level: Agent Builder, Gemini family models, custom training and deployment, function calling, grounding, and evaluation services.
• BigQuery for analytical workloads and as a grounding source: SQL fluency, partitioning and clustering, BI Engine, and the BQ-Vertex AI integration patterns.
• GCP fundamentals: IAM, VPC Service Controls, Cloud Run, Pub/Sub, Cloud Storage, Secret Manager, and the data-residency and security controls that matter in a regulated tenant.
• Vertex AI Search or equivalent retrieval layer.

Microsoft platform

• Microsoft Copilot Studio (or Power Virtual Agents) at strong level, with proven production deliveries.
• M365, Entra ID, Azure AD, Power Platform. You can reason about identity, conditional access, and tenant boundaries, not just consume them.
• Power Automate for backend flows and integrations, Power Apps for internal tooling and UI extensions.
• Microsoft Graph, Dataverse, Azure Bot Framework.

What We’re Looking For

• You’re a builder - you like to ship things that work and make people’s lives easier.
• You’re pragmatic - you know when to go fast and when to slow down.
• You’re collaborative - you work well with various stakeholders.
• You’re a self-starter - you take the initiative and get things done.
• You care about the details - especially when it comes to accessibility and user experience.

Rewards and benefits

We want to make sure that it’s possible for you to strike the right balance between your career and your private life and reward our people with a generous benefits offer and employment conditions.