Deputy Vice President, AI/ML Engineering

Job Description

Build the Intelligence Layer of Singapore's National Logistics Carrier.

From Scratch.

SingPost is executing a foundational technology reset. Legacy platforms are being sunsetted. The entire digital backbone is being rebuilt — not modernised incrementally, but redesigned from first principles for the next decade of operations.

The Intelligence Layer — the AI/ML engineering division that will make those platforms smart — is being built from zero.

• No legacy models to maintain.

• No inherited pipelines to defend.

• No prior architectural decisions to work around.

• This is a greenfield AI/ML engineering remit at national logistics scale.

We are hiring a DVP, AI/ML Engineering to own the Intelligence Layer end-to-end — to frame AI problems, write feature pipelines, build the MLOps platform, govern AI in production, and grow a team of engineers who will carry SingPost's AI capability for the next decade.

OVERALL ROLE PURPOSE

SingPost moves items daily across our national postal, logistics, and e-commerce network. This massive data volume provides the foundation for the Intelligence Layer to generate automated decisions. The data is already available. The infrastructure is being rebuilt. What is missing is the AI/ML engineering layer that converts that data into decisions, predictions, and automation that move the business forward.

The DVP, AI/ML Engineering leads the Intelligence Layer — the engineering division responsible for building, deploying, and operationalising AI and machine learning systems that generate measurable business value across logistics, finance, and operations. Better delivery ETAs. Smarter routing. Automated exception handling. Financial risk prediction. The use cases are real, the data is available, and the platform is being built to support them.

This is a greenfield remit. There is no AI/ML team to inherit, no existing models to maintain, and no technical debt to defend. The DVP will define what production AI looks like at SingPost — the standards, the architecture, the governance model, and the team.

In the early months, before the team is built, the DVP carries significant personal technical output. This requires someone who is energised by building from nothing — not someone looking for an established platform to run.

Key Responsibilities and Result Areas

Hands-On AI/ML Engineering (~30%)

• Frame every AI/ML problem in writing before a line of code is written: prediction target, success metric, data requirements, baseline — and the answer to 'what happens if this model is wrong?'

• Write production-grade feature engineering pipelines in Python: extract from BigQuery, transform, load into Vertex AI Feature Store; enforce training–serving consistency from the first pipeline

• Build and own Vertex AI Pipelines (Kubeflow SDK): full DAGs from data ingest through training, evaluation, conditional logic (skip training if drift threshold not met), and model registration

• Configure Vertex AI Online and Batch Prediction endpoints: autoscaling, traffic-split A/B testing, shadow mode — not just click through the UI, but write and own the configuration

• Implement Vertex AI Model Monitoring: skew/drift thresholds, alert triage runbooks, retraining triggers — you are the last line of defence before the business notices

• Build LLM/GenAI pipelines where they create real value: RAG design, embedding generation, retrieval chain, prompt template management, output evaluation harnesses

• Debug production failures hands-on: inspect pipeline run logs, query BigQuery prediction logs, reproduce feature pipeline failures locally — you do not wait for reports

AI Gateway & DSB Integration (~15%)

• Define and enforce the AI Gateway integration standard: all Vertex AI endpoints served through Apigee; zero direct calls from applications or business platforms — catch violations in code review, not in production

• Own the API contract between Vertex AI serving endpoints and the Apigee proxy layer: request/response schema, latency SLA, error handling, and versioning strategy

• Partner with DVP, Digital Engineering Platforms on Apigee proxy policy configuration for AI endpoints: rate limiting, auth, payload validation, response caching

MLOps Platform Ownership (~20%)

• Architect and own the MLOps platform on GCP: Vertex AI Pipelines (orchestrator), Feature Store (shared features), Model Registry (catalogue), BigQuery (training and evaluation store)

• Define and enforce model governance before a model touches production: model cards (mandatory), evaluation report sign-off, bias/fairness assessment, rollback test results, data lineage documentation

• Set the CI/CD standard for ML model artefacts: training triggered by code merge, evaluation gate before registration, automated promotion to staging, manual DVP sign-off for production

PDPA & Data Governance (~10%)

• Enforce PDPA compliance for all AI/ML data: classify training datasets, apply column-level access controls in BigQuery, ensure all training jobs run within GCP Singapore (no data egress), maintain audit logs

• Own the PDPA impact assessment for every new AI use case — before development begins, not after

AI Acceleration Programme (~10%)

• Own the Intelligence Layer's use case backlog: prioritise by business value and technical feasibility, not by what is technically interesting

• Report programme progress to VP in structured monthly updates — outcomes, not activity

• Evaluate AI tooling, foundation models, and agentic architecture patterns relevant to SingPost's logistics and financial operations context; make recommendations, not presentations

Team Building & Leadership (~15%)

• Recruit and onboard the Intelligence Layer from zero: write technical assessments, conduct interviews, make hiring decisions — starting with AI Engineer (Logistics/Finance Intelligence Squad) and ML Engineer (Predictive Operations Squad)

• Line-manage all Intelligence Layer engineers: 1:1s anchored to specific technical skill development targets, not project status updates

• Coordinate the offshore distributed engineering pod: scope data engineering tasks, review pipeline code, hold the quality bar

Requirements

• Bachelor’s or Master’s degree in Computer Science, Data Science, AI, Mathematics, or a related quantitative field

• Google Cloud Professional Machine Learning Engineer certification preferred; Professional Data Engineer certification advantageous

• 8+ years in technology roles, including:

  • 5+ years in hands-on AI/ML engineering within production environments

  • 3+ years in technical engineering leadership

• Proven experience building and operating end-to-end MLOps pipelines across training, deployment, monitoring, and retraining

• Strong Python ML engineering capability using frameworks such as scikit-learn, XGBoost, PyTorch/TensorFlow, pandas/NumPy, and FastAPI/Flask

• Deep hands-on expertise with Vertex AI, BigQuery, ML CI/CD automation, and production deployment workflows

• Experience building GenAI/LLM solutions, including RAG pipelines, embeddings, and evaluation frameworks

• Experience within complex enterprise environments; logistics, supply chain, or financial services exposure is advantageous

• Strong ability to communicate AI/ML concepts, model decisions, and technical recommendations to both technical and business stakeholders