Data & AI Architect

What matters most

• Set the enterprise target state

  The role owns Data & AI blueprints and domain roadmaps covering data mesh or lakehouse approaches, streaming, analytics, MLOps, and metadata and lineage. These must align with business strategy and regulatory obligations.

• Act as cross-domain design authority

  Review designs spanning channels, core banking, risk, finance and compliance; arbitrate decisions and manage exceptions with clear remediation timelines.

• Make standards usable in delivery

  Publish reference architectures for data modelling, eventing, the ML lifecycle, feature stores and responsible AI. Partner with delivery leads and check conformance before build, during build and before go-live.

• Embed controls in architecture

  Build data quality, lineage, retention, encryption and access controls into designs that meet privacy, model risk and operational resilience requirements across jurisdictions.

• Demonstrate value and build capability

  Track outcomes such as time-to-model, reuse and cost-to-serve through OKRs; use architecture insights to inform investment priorities while coaching architects and engineers and developing reusable implementations.

Likely challenges to think through

• Consistency versus domain autonomy

  Preparation hypothesis: Cross-domain reviews may require balancing common standards against differing needs in channels, core banking, risk, finance and compliance. Prepare a decision approach that records trade-offs, allows bounded exceptions and sets remediation dates.

• Modernisation versus delivery continuity

  Preparation hypothesis: Roadmaps for multi-cloud or hybrid data platforms may need to reconcile target-state patterns with systems already supporting batch and streaming workloads. Prepare a phased migration example showing dependencies, interoperability choices and how conformance would be checked during delivery.

• AI speed versus control evidence

  Preparation hypothesis: Real-time scoring and RAG designs may create pressure to move quickly while preserving lineage, privacy and model risk controls. Prepare to explain where controls enter the data and ML lifecycle, what evidence a design review needs and how those choices affect time-to-model.

Topics to prepare

• Enterprise data architecture trade-offs

  The role defines target-state blueprints and roadmaps for data mesh, lakehouse, analytics and hybrid or multi-cloud environments. Suggested revision exercise: compare two target-state options for a bank data domain, explaining ownership, integration and migration trade-offs.

• Batch, streaming and eventing principles

  The posting calls for batch and streaming designs, eventing standards and real-time use cases. Suggested revision exercise: explain when a banking data flow should use batch versus Kafka-based streaming, including failure handling and data-quality implications.

• Data governance and cross-jurisdiction controls

  The role must embed quality, metadata, lineage, retention, encryption and access controls while addressing privacy and operational resilience. Suggested revision exercise: trace one data product from collection to use and identify where each control belongs.

• ML lifecycle, model risk and responsible AI

  The posting covers MLOps, feature stores, model risk controls, privacy-by-design and responsible AI, with guidance on RAG and real-time scoring. Suggested revision exercise: map the controls and review points across a model’s development, deployment and ongoing operation.

• Architecture governance and measurable value

  The architect acts as a cross-domain design authority, manages dispensations and tracks outcomes such as time-to-model, reuse and cost-to-serve. Suggested revision exercise: reason through a design exception, its remediation deadline and the outcome measure that would show whether the decision worked.

Skills to revise

• Draft a target-state blueprint and roadmap

  Suggested practice: sketch a current-to-target migration for a bank data platform, showing data flows, domain boundaries, dependencies and staged decisions. This mirrors the posting’s responsibility for enterprise blueprints and domain roadmaps.

• Review a cloud and Kafka data design

  Suggested practice: annotate a batch-and-stream architecture using Kafka and one cloud platform named as an example in the posting; identify orchestration, failure recovery and conformance questions. The role requires hands-on familiarity with these technologies and design assurance.

• Specify an ML and AI reference pattern

  Suggested practice: draw a feature pipeline and real-time scoring pattern, then add a bounded RAG/vector-store component and its model-risk checkpoints. These are solution patterns the posting says the architect will guide, not a requirement to use every pattern in one solution.

• Build a lineage and controls checklist

  Suggested practice: create a review checklist linking a data product’s sources, transformations and consumers to catalog/lineage records, quality checks, retention and access controls. This reflects the stated governance and assurance responsibilities.

• Run a cross-domain design review

  Suggested practice: write a one-page decision record for a proposed design spanning channels, risk and compliance; state the trade-off, stakeholders, any dispensation and a remediation timeline. The posting explicitly calls for adjudicating decisions and influencing business, technology, risk and compliance.

• Map your evidence to the role

  Prepare a one-page evidence map for the stated architecture experience, including leadership of data and AI at scale, platform modernisation, MLOps, hybrid or multi-cloud delivery, and cross-domain influence. Use only examples you can substantiate.

• Build a modernisation walkthrough

  Select one relevant project, if available, and prepare a diagram and brief account of the starting platform, target architecture, migration decisions, your role, trade-offs, and measured outcomes.

• Practise a target-state design

  As a preparation exercise, sketch a bank-wide data and AI blueprint and phased roadmap covering batch and streaming flows, lakehouse or data-mesh choices, analytics, MLOps, and metadata and lineage. Explain assumptions and regulatory trade-offs rather than presenting it as OCBC’s architecture.

• Prepare a design-review decision

  Draft a sample architecture decision record for a cross-domain proposal: compare options, identify where a standard applies, and show how you would document an exception, owner, and remediation timeline. The role acts as design authority across bank programmes.

• Trace controls through an AI use case

  Create a simple design walkthrough, such as real-time scoring or RAG, showing where data quality, lineage, retention, encryption, access, privacy, model-risk, responsible-AI, and resilience controls would be checked before and during delivery and before launch.

• Show how you drive adoption and value

  Prepare one example, if available, of coaching architects or engineers and spreading a reusable pattern. Pair it with a concise explanation of how you would track outcomes such as time-to-model, reuse, or cost-to-serve through OKRs.

• Write questions for OCBC

  Prepare questions about the current target-state roadmap, design-review and exception process, cross-jurisdiction control priorities, and how architecture outcomes are measured. These are questions to verify, not assumptions about OCBC’s current practices.