Job Description

Empowering Africa’s tomorrow, together…one story at a time.

Absa Group’s Chief Data Analytics and Applied AI Office (CDAIO) requires a highly skilled and production-focused AI Engineer (Data Scientist) to join the AI Solution Engineering Squad. This role plays a key role in the development, deployment, and operationalisation of AI capabilities such as machine learning (ML), artificial intelligence (AI), generative AI, and agentic AI models. The capabilities must enable the CDAIO to fulfil its mandate as steward of the bank’s AI capabilities through the end-to-end delivery of the AI solutions, governance and acceptable use in service of the bank’s strategic and commercial objectives.

This role provides AI solution engineering across four business units (CIB, PPB, BB, and AR) and ten countries.

The successful candidate will design, develop, deploy, and govern production-grade ML, generative AI, and agentic AI solutions that deliver measurable, auditable business value across the Group. This role demands an exceptional combination: deep applied AI engineering capability, rigorous model risk management discipline embedded from the start of each delivery, responsible AI as a first-class engineering practice built into the CI/CD pipeline, and sufficient quantitative finance literacy to design solutions that interact safely with Absa's regulatory capital, credit, and compliance frameworks.

This is a delivery role, not a research role. The successful candidate treats explainability, fairness, and audit-readiness as engineering requirements, not afterthoughts, and engages proactively with Group Model Risk, Risk, and Compliance as core delivery partners rather than downstream sign-off functions. The role includes mentoring junior data scientists across the Group and contributing to the CDAIO's AI practitioner community and responsible AI governance forums.

KEY FOCUS AREAS

• AI and ML Solution Delivery Production-grade design, development, and deployment of ML, generative AI, and agentic AI solutions across Absa's four business units and ten countries, using the CDAIO's enterprise AI platform stack.
• Agentic AI Engineering End-to-end design and operation of agentic AI systems including multi-agent orchestration, human-in-the-loop workflows, agent evaluation frameworks, and safety guardrails for regulated financial services contexts.
• Model Risk Management and Regulatory Compliance: Full MRM lifecycle accountability: model documentation, explainability, bias and fairness testing, formal MRM sign-off, and ongoing monitoring, in compliance with Absa's Model Risk Management framework, POPIA, NCA, and applicable AI regulation across Absa's footprint.
• Responsible AI as a Delivery Discipline Embedding responsible AI controls into the CI/CD pipeline, producing model cards and datasheets, conducting responsible AI impact assessments for high-risk use cases, and participating in AI ethics governance as the engineering voice.
• MLOps and Model Lifecycle Governance Champion-challenger testing, automated drift monitoring, model retirement, Group model inventory management, and quarterly model performance reporting to the Head of AI Technology and Group Model Risk.
• Financial Domain Application Applied AI in credit risk, fraud, AML, customer personalisation, KYC, and regulatory reporting, with working knowledge of how model outputs interact with International Financial Reporting Standards (IFRS9), Basel III capital frameworks, and South African consumer credit regulation.

ACCOUNTABILITIES

AI and ML Solution Design and Delivery

• Design, develop, test, and deploy production-grade ML, generative AI, and agentic AI solutions using the CDAIO's AI platform stack: Databricks, AWS Bedrock, Microsoft Azure AI Foundry, and Hugging Face.
• Design and develop reusable AI blueprints — standardised solution patterns for common financial services use cases — enabling consistent, accelerated delivery across business units and materially reducing rework.
• Translate complex business problems into structured AI solution designs: model selection rationale, feature engineering approaches, training data requirements, and deployment architecture in collaboration with BU technology and product teams.
• Build model pipelines using enterprise MLOps frameworks, ensuring end-to-end auditability, scalability, versioning, and performance tracking in production.
• Develop and implement AI guardrails in accordance with Group security standards, architecture policies, and the CDAIO's AI Responsible Use Governance framework — including output validation, input sanitisation, and human-in-the-loop controls for high-risk decisions.
• Partner with business unit and Group function stakeholders (Risk, Compliance, Legal, Finance) to identify, prioritise, design, and deploy AI solutions that generate demonstrable and tracked business value.
• Monitor and maintain model performance, retraining and optimising as required in dynamic banking environments; stay current with advances in foundation models, multi-agent systems, and AI governance frameworks.

Second Capability

• Design and build production-grade agentic AI systems using applications such as LangGraph, AutoGen, CrewAI, or AWS Bedrock Agents, with clearly defined orchestrator and subagent role boundaries and documented agent interaction protocols.
• Implement agent memory architectures; short-term context management, long-term memory stores, and episodic retrieval that are appropriate to the task type and regulatory risk level of each deployment.
• Design and implement human-in-the-loop (HITL) workflows for agentic systems operating in regulated decision contexts, with appropriate escalation triggers, approval gates, and audit trails required before production deployment.
• Build and maintain agent evaluation frameworks covering task completion rates, safety boundary testing, output quality scoring, and failure mode analysis, ensuring agentic systems meet defined performance and safety thresholds before and after deployment.
• Implement agent safety guardrails: output validation schemas, action whitelisting, tool-calling constraints, and circuit-breaker patterns preventing unintended autonomous action.
• Develop RAG architectures using vector databases (OpenSearch, Pinecone, FAISS, Milvus) and AI orchestration frameworks (LangChain, LlamaIndex), ensuring retrieval quality, latency, and data governance standards are met.
• Design and deploy AI solutions with awareness of data localisation, cross-border data transfer, and country-specific regulatory requirements across Absa's ten operating countries.

Model Risk Management and Regulatory Compliance

• Prepare and maintain comprehensive model documentation packages for all production models; development methodology, validation plans, performance benchmarks, explainability outputs, and ongoing monitoring frameworks, in the format required by Absa's Group Model Risk Management (MRM) function.
• Ensure all models achieve formal MRM sign-off before production deployment. Engage proactively with Group Model Risk throughout the development lifecycle, not solely at the point of submission.
• Implement model explainability as a first-class engineering requirement using SHAP, LIME, DALEX, or equivalent techniques appropriate to each model type and regulatory risk level; explainability outputs must be produced as part of the standard development pipeline, not added post-hoc.
• Conduct bias and fairness testing for all customer-facing and credit-decision models under obligations arising from the National Credit Act (NCA), POPIA, and Absa's Group Fairness Policy, documenting results, remediation actions, and residual risk in the model record.
• Manage model change governance through Absa's model inventory and change management process, ensuring material model changes trigger re-validation and updated MRM sign-off before deployment.
• Maintain current awareness of AI regulatory frameworks across Absa's footprint: EU AI Act obligations applicable to Absa's European-linked operations, SARB guidance on AI in financial services, and country-specific AI and data regulations across ten operating countries.

Responsible AI as a Delivery Accountability

• Integrate responsible AI controls directly into the CI/CD pipeline for all model development, automated bias testing gates, model card generation, fairness metric tracking, and explainability report production, so that responsible AI is embedded in delivery, not applied as a post-build compliance overlay.
• Produce and maintain model cards and datasheets for all production models: training data provenance, known limitations, intended and out-of-scope use cases, bias testing results, and recommended monitoring thresholds.
• Conduct responsible AI impact assessments for all high-risk AI use cases in collaboration with Group Risk and the CDAIO's AI Responsible Use Governance capability, covering potential harms to customers, third parties, and the Group, before development commences.
• Participate in the CDAIO's AI ethics and responsible AI governance forums as the engineering voice: translating technical realities into governance language and ensuring governance decisions are grounded in operational reality.
• Develop and enforce AI solution guardrails aligned to the CDAIO's AI Responsible Use Policy: data minimisation principles, consent management for training data, and transparency requirements for AI-assisted customer decisions.

MLOps and Model Lifecycle Governance

• Build and maintain champion-challenger testing frameworks for all production models, ensuring new model versions are validated against the incumbent before promotion to production and that selection rationale is documented in the model record.
• Implement automated model drift monitoring with threshold-triggered retraining pipelines; drift thresholds to be defined in consultation with Group Model Risk and the relevant business unit, all retraining events to be logged and reviewed.
• Own the model retirement and decommissioning process for all CDAIO portfolio models, ensuring retired models are removed from Absa's Group model inventory with appropriate documentation and that downstream dependencies are managed.
• Maintain the CDAIO's production model inventory, contributing to Absa's Group model register with current performance baselines, monitoring dashboards, and escalation contacts recorded for every active model.
• Produce quarterly model performance review reports for the Head of AI Technology and Group Model Risk: drift trends, champion-challenger outcomes, retraining history, and model risk events.

Financial Domain Application

• Apply AI and ML to Absa's priority financial services use cases: credit risk modelling (PD, LGD, EAD), fraud detection and AML (transaction monitoring, network graph analysis, typology-based detection), customer personalisation and next-best-action, KYC automation, and regulatory compliance analytics.
• Demonstrate working knowledge of how ML model outputs interact with regulatory capital calculations, IFRS 9 expected credit loss provisioning, stress testing requirements, and internal risk reporting, engaging with quantitative analysts and actuaries where model design choices have capital or reporting implications.
• Design AI solutions with full awareness of the operational and regulatory context of each Absa business unit (CIB, PPB, BB, and AR) accounting for the differing risk profiles, data environments, and regulatory obligations of each.
• Collaborate with BU data science and analytics teams to formalise AI use case requirements, building solutions that integrate with existing BU data products, reporting infrastructure, and technology platforms.

People and Capability Development

• Provide structured technical mentoring to junior data scientists and AI engineers across the Group: model selection, experimentation protocols, responsible AI practices, MLOps disciplines, and deployment best practices.
• Conduct peer reviews of model development work across the squad, maintaining quality standards for code, model documentation, explainability outputs, and MRM submission materials.
• Actively contribute to the CDAIO's AI practitioner community through internal knowledge-sharing sessions, written documentation, and structured capability-building programmes.
• Maintain comprehensive documentation of all AI solutions, architectural decision records, training data catalogues, model versioning histories, and deployment runbooks, ensuring institutional knowledge is independent of individual team members and contractor dependencies are actively mitigated.

QUALIFICATIONS AND EXPERIENCE

Education/ Qualification:

• Postgraduate degree (Masters or PhD) in a quantitative discipline such as Computer Science, Data Science, Mathematics, Statistics, Engineering, or equivalent. Candidates with demonstrated production AI delivery experience are preferred over academic credentials alone.
• Certification in the following are advantageous:
  • Cloud AI Certification - Databricks Certified Machine Learning Professional, AWS Certified Machine Learning Specialty, or Microsoft Azure AI Engineer Associate
  • Responsible AI - Certified in AI governance, ethics, or responsible AI (IAPP, IEEE, or institutional equivalent)
  • Financial Domain - Structured training or qualification in at least one of: credit risk, AML/fraud analytics, IFRS 9, or regulatory compliance analytics.

Work Experience:

• 8-12 years of progressive leadership experience in Applied AI (Data Science)’ with production delivery of ML, generative AI, or agentic AI solutions in an enterprise environment, preferably financial services or a similarly regulated industry.
• 2-3 years experience in the following:
  • Generative AI Production- Demonstrated production deployment of at least one LLM, RAG, or agentic AI solution beyond proof-of-concept or internal tooling, with documented business value and governance compliance.
  • Model Risk Management - Preparing model documentation packages and engaging with an MRM function through formal sign-off: explainability, bias and fairness testing, validation, and ongoing monitoring submissions.
  • Responsible AI Delivery- Integrating responsible AI controls into CI/CD pipelines; producing model cards and datasheets; conducting responsible AI impact assessments in a regulated environment.
  • MLOps and Model Lifecycle- Champion-challenger frameworks, automated drift monitoring, threshold-triggered retraining, and model retirement in production MLOps environments.
  • Agentic AI Engineering- Design and deployment of multi-agent systems, HITL workflows, agent evaluation frameworks, and safety guardrails in a production or near-production regulated context.
  • Financial Services Domain- Applied AI in at least two of: credit risk (PD/LGD/EAD), fraud/AML (transaction monitoring, typology detection, graph ML), customer personalisation, KYC automation, or IFRS 9 provisioning analytics.
  • Mentoring and Peer Review-Mentoring or coaching junior data scientists including peer review of model development, documentation quality, and MRM submission materials.
• Advantageous:
  • Pan-African Deployments- Working across multiple African jurisdictions with awareness of data localisation and AI regulatory requirements.

Knowledge and Skills:

• Applied AI and ML Expert-level: Python, Scikit-learn, TensorFlow, PyTorch, Hugging Face Transformers; classical ML, LLMs, GANs, VAEs, and their application to financial services use cases.
• Generative AI and RAG: LLM architectures, RAG pipeline design, fine-tuning, prompt engineering, structured output generation; vector databases (OpenSearch, Pinecone, FAISS, Milvus); LangChain and LlamaIndex.
• Agentic AI Engineering: LangGraph, AutoGen, CrewAI, or AWS Bedrock Agents; agent memory architectures; tool-calling API design; HITL workflow design; agent evaluation frameworks; safety guardrail implementation.
• MLOps: MLflow, Kubeflow, Docker, Kubernetes, Airflow, CI/CD for ML; champion-challenger frameworks; drift monitoring; automated retraining; model versioning and retirement.
• Model Risk Management: MRM framework requirements; model documentation standards; explainability techniques (SHAP, LIME, DALEX); bias and fairness testing (Fairlearn, AI Fairness 360); model change governance and validation processes.
• Responsible AI: Responsible AI impact assessment methodology; model card and datasheet production; fairness metrics; data minimisation; consent management for training data; AI ethics governance forums.
• Financial Services Domain: Credit risk modelling (PD/LGD/EAD); fraud detection and AML typologies; IFRS 9 provisioning analytics; regulatory capital interaction; NCA and POPIA obligations; SARB AI guidance.
• Enterprise AI Platform Stack: Databricks AI (Unity Catalog, Delta Lake, Workflows); AWS Bedrock and SageMaker; Microsoft Azure AI Foundry; Hugging Face enterprise; enterprise data pipelines and data governance standards.
• AI Regulatory Frameworks: EU AI Act obligations; SARB AI guidance; POPIA; country-specific AI and data regulations across Absa's ten African operating countries; model risk regulatory expectations (SR 11-7 equivalent principles).
• Data Architecture and Governance Enterprise data pipeline design; data lineage; access control; audit trail requirements; governance standards for AI solutions in a regulated financial services environment.
• Agile Delivery: Experimentation discipline; sprint planning; backlog management; test-driven development for ML; continuous delivery practices in a self-directed squad operating under governance obligations.

Education

Bachelor's Degree: Information Technology

Absa Bank Limited is an equal opportunity, affirmative action employer. In compliance with the Employment Equity Act 55 of 1998, preference will be given to suitable candidates from designated groups whose appointments will contribute towards achievement of equitable demographic representation of our workforce profile and add to the diversity of the Bank.

Absa Bank Limited reserves the right not to make an appointment to the post as advertised