New opportunity

[LTA-TRO] AI SYSTEM ARCHITECT — TRANSPORT AI PROGRAMME

LTA Land Transport Authority · Singapore

About this role

The AI System Architect will own the technical architecture of the Land Transport Authority’s Transport AI Programme in Singapore. Reporting to the Programme Director, the architect will ensure AI systems are scalable, secure, interoperable and ready for deployment in critical public infrastructure, while connecting programme strategy with engineering delivery.

Responsibilities

Define end-to-end architecture spanning model serving, data platforms, integration layers, AI safety infrastructure and observability. Set patterns and standards for generative AI, agentic workflows, RAG, computer vision pipelines and shared AI platforms.
Review industry-partner solutions and conduct technical due diligence against security, privacy, resilience and operational-readiness standards. Design infrastructure for automated evaluations, risk-posture dashboards, model drift monitoring and deployment playbooks.
Collaborate with data governance, security and policy teams on personal data protection, prompt-injection defence and responsible AI. Mentor AI developers and infrastructure engineers, serve as the primary technical interface with external partners, and contribute to whole-of-government AI platform discussions.

Requirements

Knowledge of Computer Science, Computer Engineering or a related field; at least 3–6 years of relevant professional experience in software or systems architecture, including at least 3 years focused on AI or data-intensive systems. Candidates may be considered at intermediate or senior level based on experience.
Deep familiarity with LLM serving infrastructure, vector databases, orchestration frameworks and MLOps tooling, plus cloud-native architecture, API design and enterprise integration patterns. Experience designing systems for regulated or high-availability environments is important.
Strong communication and cross-functional collaboration skills, including the ability to explain technical concepts to engineers and senior non-technical stakeholders, translate operational data needs into engineering solutions and work in a fast-moving Agile environment. Experience in government, transport or critical infrastructure is an advantage.

Shortlisted candidates may need to complete a medical declaration and/or further assessment.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

AI systems architectureLLM servingVector databasesOrchestration frameworksMLOpsCloud-native architectureAPI designEnterprise integrationGenerative AIAI agentsRAGComputer visionModel servingData platformsObservabilityAI evaluationModel drift monitoringAI safetyData governancePersonal data protectionPrompt-injection defenceResponsible AISecurity architectureTechnical due diligenceTechnical leadershipStakeholder communicationAgile

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Own the end-to-end AI architecture
The role defines and maintains architecture across model serving, data platforms, integration, AI safety infrastructure and observability. It must be scalable, secure, interoperable and fit for deployment in critical public infrastructure.
Set reusable AI platform standards
The architect establishes patterns for generative AI, agentic workflows, RAG, computer vision pipelines and shared platforms, while guiding engineering decisions across implementations.
Make AI systems operationally assessable
The posting calls for infrastructure supporting automated evaluations, risk-posture dashboards, model drift monitoring and deployment playbooks. Operational readiness is also a criterion in partner architecture reviews.
Review partner solutions and embed safeguards
The architect leads technical due diligence against security, privacy, resilience and operational-readiness standards, and works with governance, security and policy teams on personal data protection, prompt-injection defence and responsible AI.
Connect strategy, delivery and external platforms
Reporting to the Programme Director, the architect guides and mentors engineers, acts as the primary technical interface with industry partners, and contributes to whole-of-government platform discussions where relevant.

Likely challenges to think through

Standardisation versus workload-specific needs
Likely preparation challenge, not a reported LTA problem: shared standards must accommodate different needs in RAG, agentic and computer vision systems without fragmenting the platform. Prepare an example of deciding what belongs in a common architecture and what should remain workload-specific.
Deployment pace versus assurance
Likely preparation challenge: in critical infrastructure, advancing an AI system toward deployment may compete with time needed for evaluation, privacy and security review, and resilience testing. Prepare to explain risk-based release criteria, monitoring and a response to drift or failed evaluations; these are suggested discussion points, not stated LTA procedures.
Partner fit versus architectural control
Likely preparation challenge: an industry-partner solution may offer useful capabilities yet fall short on integration, observability or operational readiness. Prepare a due-diligence approach that makes trade-offs explicit and distinguishes issues to resolve from reasons not to adopt.
Technical decisions across diverse stakeholders
Likely preparation challenge: the architect must translate the Programme Director’s strategy and operational data needs into engineering decisions while engaging governance teams, partners and non-technical leaders. Prepare a concise architecture rationale that explains trade-offs differently for technical and non-technical audiences.

Practice your answers

Technical practice

Suggested practice: How would you define the end-to-end architecture for a shared transport AI platform?

What to coverThe role owns architecture across model serving, data platforms, integration layers, AI safety and observability. Inferred preparation suggestion: explain component boundaries, shared standards and how you would balance scalability, security and interoperability.

Suggested practice: How would you design LLM serving for a high-availability environment?

What to coverThe posting calls for deep familiarity with LLM serving and experience in high-availability environments. Inferred preparation suggestion: discuss capacity, latency, failure handling and the trade-offs involved in keeping a service operational.

Suggested practice: What architectural choices would you make for a RAG system?

What to coverRAG patterns and vector databases are explicitly named in the posting. Inferred preparation suggestion: cover document ingestion, retrieval quality, access controls and how you would evaluate whether retrieved material supports an answer.

Suggested practice: How would you set standards for agentic workflows?

What to coverEstablishing patterns for agentic workflows is a stated responsibility. Inferred preparation suggestion: address orchestration, limits on agent actions, human oversight and ways to observe or stop unsafe behaviour.

Suggested practice: How would you take a computer vision pipeline from design to operational deployment?

What to coverThe role includes setting patterns for computer vision pipelines in a critical public infrastructure context. Inferred preparation suggestion: describe data flow, inference integration, performance evaluation and how the pipeline would handle degraded inputs or failures.

Suggested practice: How would you integrate AI services with existing enterprise systems?

What to coverThe posting names API design, enterprise integration patterns, cloud-native architecture and cross-government platform alignment. Inferred preparation suggestion: explain interface contracts, data boundaries, versioning and trade-offs between a shared service and a system-specific integration.

Suggested practice: How would you design for personal data protection and prompt-injection defence?

What to coverThe architect works with governance, security and policy teams on personal data protection, prompt-injection defence and responsible AI. Inferred preparation suggestion: describe layered controls and how you would test their effectiveness without treating any single control as sufficient.

Suggested practice: How would you evaluate and monitor an AI system after deployment?

What to coverThe role includes automated evaluations, risk-posture dashboards and model drift monitoring. Inferred preparation suggestion: distinguish pre-deployment evaluation from ongoing monitoring, and explain what evidence would trigger investigation or corrective action.

Suggested practice: How would you conduct technical due diligence on an industry partner’s AI solution?

What to coverThe posting requires architecture reviews against security, privacy, resilience and operational-readiness standards. Inferred preparation suggestion: explain what evidence you would request, how you would test important claims and how you would document unresolved risks.

Suggested practice: What would you include in an AI deployment playbook?

What to coverDeployment playbooks, observability and operational readiness are stated responsibilities. Inferred preparation suggestion: cover release checks, monitoring, incident response and rollback, including the trade-off between rapid delivery and controlled deployment.

Behavioural practice

Suggested practice: Tell me about a time you translated a strategic goal into engineering decisions.

What to coverThe role bridges the Programme Director’s vision and day-to-day engineering delivery. Inferred preparation suggestion: use a real situation, the architectural decisions and communication you personally led, and the resulting outcome.

Suggested practice: Tell me about a time you reconciled operational needs with security or governance concerns.

What to coverThe posting calls for collaboration across operational, data governance, security and policy stakeholders. Inferred preparation suggestion: describe the conflicting needs, how you worked through them and what the agreed solution achieved.

Suggested practice: Tell me about a time you mentored engineers through an architectural decision.

What to coverTechnical leadership and mentorship of AI developers and infrastructure engineers are explicit duties. Inferred preparation suggestion: explain the decision context, how you supported others without simply dictating an answer, and the result.

Suggested practice: Tell me about a time you challenged an external partner’s technical proposal.

What to coverThe architect leads partner architecture reviews and serves as a primary technical interface. Inferred preparation suggestion: give the situation, the evidence and discussion you used, and the outcome without implying a review you have not performed.

Suggested practice: Tell me about a time you aligned technical and non-technical stakeholders during fast-moving delivery.

What to coverThe posting emphasises clear communication with engineers and senior non-technical stakeholders in a collaborative Agile environment. Inferred preparation suggestion: describe the differing perspectives, how you explained the trade-offs and what changed as a result.

Sharpen your knowledge

Topics to prepare

End-to-end AI platform architecture
The posting assigns ownership of architecture spanning model serving, data platforms, integration layers, AI safety and observability. Suggested exercise: trace a hypothetical AI request from data ingestion through serving and monitoring, explaining the purpose and boundaries of each layer.
AI workload patterns
LTA asks for architectural standards for generative AI, agentic workflows, RAG, computer vision and shared AI platforms. Suggested exercise: compare where each pattern needs distinct data flows, controls and operational safeguards; include the roles of vector databases and orchestration frameworks.
Cloud-native integration and resilience
The posting calls for cloud-native architecture, API design and enterprise integration patterns, with experience in regulated or high-availability environments. Suggested exercise: reason through how an integration should behave during an upstream outage and how its interfaces could remain interoperable.
Security, privacy and responsible AI
The architect works with governance, security and policy teams on personal data protection, prompt-injection defence and responsible AI. Suggested exercise: map likely trust boundaries in a hypothetical RAG flow and explain which concerns require cross-functional decisions.
Evaluation and operational readiness
The role includes infrastructure for automated evaluations, risk-posture dashboards, model drift monitoring and deployment playbooks. Suggested exercise: distinguish what each detects or supports, and explain how the findings could inform a deployment decision.

Skills to revise

Draw and defend an architecture
Suggested exercise: create a one-page design for a hypothetical transport AI service showing serving, data, APIs, safety controls and observability; then defend its scalability and interoperability choices. This practises the end-to-end architecture ownership described in the posting, not a specified LTA project.
Conduct a partner architecture review
Suggested exercise: draft review questions and an evidence checklist covering security, privacy, resilience and operational readiness for a hypothetical industry-partner solution. The posting explicitly assigns technical due diligence and partner-facing reviews.
Specify evaluation and monitoring flows
Suggested exercise: outline what an automated evaluation records, how drift would be surfaced, and what information a risk-posture dashboard and deployment playbook would need. LTA explicitly names these infrastructure responsibilities and MLOps familiarity.
Communicate trade-offs across audiences
Suggested exercise: explain the same architecture decision first to engineers, then to a senior non-technical stakeholder, including its operational consequence and an unresolved trade-off. The posting stresses technical leadership, cross-functional collaboration and clear communication.

Your preparation checklist

Prepare an end-to-end architecture walkthrough
The role owns architecture across model serving, data platforms, integrations, AI safety and observability. If you have a relevant project, prepare a diagram explaining your decisions, trade-offs, deployment constraints and outcomes.
Practise a transport AI design exercise
As interview preparation, sketch a hypothetical RAG or computer-vision service for transport operations. Show its data flow, APIs, scaling approach, interoperability and failure handling; do not present the scenario as an LTA project.
Assemble evidence for platform expertise
Prepare brief examples showing your work with LLM serving, vector databases, orchestration, MLOps, cloud-native architecture, API design and enterprise integration. Be ready to explain what you designed versus what you used.
Map safety and operational controls
Prepare a one-page proposal for automated evaluations, drift monitoring, risk-posture reporting and deployment playbooks, including how you would address personal data protection, prompt injection and responsible AI.
Rehearse a partner architecture review
The role leads due diligence on industry-partner solutions. Draft a review checklist covering security, privacy, resilience and operational readiness, then practise explaining how you would resolve a significant finding.
Prepare two stakeholder-level explanations
The posting calls for communication with engineers and senior non-technical stakeholders. Practise explaining the same architecture decision at both levels, including how operational data needs become engineering requirements.
Bring questions about architectural ownership
Ask how the Programme Director, engineering teams, industry partners and cross-government platforms coordinate decisions, and what operational-readiness expectations would guide the architect's priorities.

Availability last checked: 4 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in Singapore. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…