New opportunity

AI Engineer, Data Infra

Nanyang Technological University · Singapore

About this role

AI Singapore (AISG) seeks an AI Engineer for its Platform team within AI Products. Hosted at Nanyang Technological University under the VP (Artificial Intelligence & Digital Economy)’s office, the role builds and scales data infrastructure for organisation-wide large language model (LLM) training, fine-tuning, evaluation and retrieval-augmented generation (RAG). The engineer contributes to the architecture and reliability of data storage, pipelines and compute infrastructure for large-scale AI workloads.

Responsibilities

Develop scalable data architecture; select and manage the data technology stack; automate data transfers and backups; and support data cataloging and discovery.
Design and maintain batch and streaming pipelines using tools such as AWS Glue and Amazon EMR.
Govern organisation-level IAM across AWS Organizations, including service control policies, cross-account roles and permission boundaries, to enforce least-privilege access. Improve data access patterns for performance and cost.
Build analytics and dashboards using tools such as Amazon Athena and QuickSight to show stakeholders platform usage and costs. Use AI tools appropriately in daily engineering and develop internal AI-assisted tools that reduce operational effort.

Requirements

A degree in Computer Science, Information Technology or equivalent, and at least 2–4 years of data infrastructure, platform or systems engineering experience operating production systems at scale.
Strong knowledge of distributed data systems, object storage, data lakes, distributed file systems, vector databases and pipeline tools such as Apache Spark, Apache Airflow, Ray or Dagster. Working knowledge of data access control and orchestration.
Familiarity with cloud organisational structures, multi-account or multi-project setups, organisation-level IAM and billing or cost allocation. Hands-on experience running workloads across cloud providers using infrastructure as code, containers, orchestration and managed compute, storage and networking services; examples include Terraform, Docker and Kubernetes.
Python and SQL skills, demonstrated use of tools such as Claude, Copilot or Cursor for engineering work, and strong communication and collaboration. The role calls for balancing platform ownership with usability, self-service and adoption.

Preferred

Experience with LLM training datasets such as Common Crawl; C, C++, Rust or Go; or contributions to open-source AI/ML projects.

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

Skills for this role

Data architectureData infrastructureData pipelinesBatch processingStream processingDistributed data systemsObject storageData lakesDistributed file systemsVector databasesData orchestrationData governanceData access controlData catalogingAWS GlueAmazon EMRAWS OrganizationsAWS IAMService control policiesAmazon AthenaAmazon QuickSightApache SparkApache AirflowRayDagsterTerraformInfrastructure as codeDockerKubernetesPython work_mode

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

Scale data architecture for LLM workloads
The role develops the data architecture and selects the technology stack for organisation-wide training, fine-tuning, evaluation and RAG, with responsibility for reliable storage and supporting compute infrastructure.
Operate dependable data pipelines
The posting calls for batch and streaming pipelines, automated data transfers and backups, and support for data cataloging and discovery. AWS Glue and Amazon EMR are named as example tools.
Govern access across cloud accounts
A stated responsibility is organisation-level IAM across AWS Organizations, including service control policies, cross-account roles and permission boundaries to enforce least privilege.
Make usage and cost visible
The engineer is expected to improve data access patterns for performance and cost, and provide stakeholders with platform usage and cost dashboards using tools such as Athena and QuickSight.
Build usable, AI-assisted platform tools
The posting emphasizes self-service and adoption by engineers and researchers, alongside appropriate use of AI tools in engineering and internal AI-assisted tools that reduce manual operations work.

Likely challenges to think through

Serving different LLM workloads with one platform
Preparation hypothesis: training, evaluation and RAG may favor different storage and access patterns, making stack choices a trade-off between scale, reliability and usability. Prepare to explain how you would compare options against workload needs and validate a design before adopting it.
Keeping self-service compatible with least privilege
Preparation hypothesis: cross-account access can make data easy to use or tightly controlled, but doing both requires careful policy design. Prepare to walk through how you would scope permissions, test access boundaries and make approved access practical for users.
Optimizing performance without obscuring cost
Preparation hypothesis: faster data access and compute can increase spending, while cost controls can slow workloads. Prepare to describe what usage and cost measures you would expose to stakeholders and how you would evaluate an optimization against both.
Reducing manual work without trusting AI output blindly
Preparation hypothesis: AI-assisted operations may save time but introduce incorrect code or operational changes. Prepare to distinguish suitable uses, such as drafting or debugging, from changes that require review, testing and human judgment.

Practice your answers

Technical practice

How would you design data storage for LLM training, evaluation, and RAG workloads?

What to coverThe role covers scalable storage for these workloads and calls for knowledge of object storage, data lakes, distributed file systems, and vector databases. Preparation suggestion: explain which access patterns each supports and how you would weigh scale, reliability, and cost.

How would you design and maintain both batch and streaming data pipelines?

What to coverThe posting names batch and streaming pipelines, with AWS Glue and Amazon EMR as examples. Preparation suggestion: compare processing needs, orchestration choices, and how you would detect and recover from failed runs.

How would you automate data transfers and backups while keeping data discoverable?

What to coverTransfer and backup automation, cataloging, and discovery are stated responsibilities. Preparation suggestion: discuss validation, recovery, and metadata that would help engineers find and understand datasets.

How would you enforce least-privilege access across AWS accounts?

What to coverThe role includes organisation-level IAM across AWS Organizations, including service control policies, cross-account roles, and permission boundaries. Preparation suggestion: explain the distinct purpose of each control and how you would check that access works without becoming overly broad.

How would you improve a costly or slow data access pattern?

What to coverReviewing access patterns for performance and cost is an explicit responsibility. Preparation suggestion: describe what usage and query evidence you would inspect, possible changes, and how you would measure the result.

How would you give stakeholders visibility into platform usage and cost?

What to coverThe posting calls for analytics and dashboards using tools such as Amazon Athena and QuickSight. Preparation suggestion: explain how you would define useful measures, account for cost allocation, and make the figures understandable to stakeholders.

How would you operate data workloads across cloud environments?

What to coverThe requirements mention multi-account or multi-project structures, infrastructure as code, containers, orchestration, and managed cloud services. Preparation suggestion: discuss repeatable deployment, access boundaries, and the trade-offs between managed services and operational control.

How would you investigate a production data pipeline that has become unreliable?

What to coverThe role requires experience operating production systems at scale and contributing to pipeline reliability. Preparation suggestion: walk through how you would establish impact, inspect failures and dependencies, restore service, and prevent recurrence.

Where would you use AI tools in engineering work, and where would you avoid relying on them?

What to coverThe posting mentions AI tools for code generation, review, debugging, and documentation, plus internal AI-assisted tools that reduce manual operations. Preparation suggestion: give criteria for verifying outputs, protecting sensitive data, and deciding whether an internal tool saves meaningful effort.

How would you use Python and SQL to investigate a data quality or pipeline issue?

What to coverPython and SQL skills, along with reading other people's code, are stated requirements. Preparation suggestion: explain how you would trace transformations, check records or aggregates, and test a fix without assuming the first discrepancy identifies the root cause.

Behavioural practice

Tell me about a time you balanced platform controls with user self-service.

What to coverThe role explicitly balances architecture, cost, and governance with usability and adoption. Preparation suggestion: use a real situation, your actions and trade-offs, and the resulting effect on users and platform operation.

Describe a time you worked with others to resolve a production infrastructure problem.

What to coverProduction operation, reliability, communication, and collaboration are relevant to this posting. Preparation suggestion: give the situation, your specific actions and coordination, and the result, distinguishing your contribution from the team's.

Tell me about a time you explained a technical or cost trade-off to stakeholders.

What to coverThe role includes technology selection and stakeholder visibility into platform usage and cost. Preparation suggestion: describe the decision context, how you communicated options, and what happened after the decision.

Describe a time you changed a tool or workflow to reduce repetitive engineering work.

What to coverThe posting calls for continuous improvement and internal AI-assisted tools that reduce manual effort. Preparation suggestion: use a real situation, the change you made, how you checked its usefulness, and the result; do not claim AI was involved if it was not.

Tell me about a time you helped others adopt a platform capability.

What to coverThe role emphasizes building tools engineers and researchers want to use, including self-service and adoption. Preparation suggestion: explain the users' initial needs, how you incorporated feedback, and what evidence showed whether the capability was adopted.

Sharpen your knowledge

Topics to prepare

Distributed storage for LLM workloads
Posting: The role supports training, fine-tuning, evaluation and RAG, and calls for knowledge of object storage, data lakes, distributed file systems and vector databases. Revision exercise: Sketch where each storage type fits in those workloads and explain the reliability, access and cost trade-offs.
Batch, streaming and orchestration
Posting: The engineer designs batch and streaming pipelines and needs knowledge of pipeline tools and data orchestration. Revision exercise: Trace a dataset from ingestion to availability, explaining scheduling, retries, data validation and when streaming is preferable to batch.
Organisation-level access control
Posting: The role governs IAM across AWS Organizations using service control policies, cross-account roles and permission boundaries. Revision exercise: Draw a two-account access model and explain what each control can restrict under least privilege.
Cloud operations and cost allocation
Posting: Requirements include production systems at scale, workloads across cloud providers, multi-account or multi-project structures, and billing or cost allocation. Revision exercise: Walk through an infrastructure change and identify its reliability, access and cost implications across accounts.
Platform ownership and self-service
Posting: The engineer must balance architecture, cost and governance with usability, self-service and adoption by engineers and researchers. Revision exercise: Outline how you would evaluate a proposed data-stack tool against those competing needs; this is a preparation exercise, not a stated selection process.

Skills to revise

Build a small data pipeline
Posting: Python, SQL, batch and streaming pipelines are relevant; AWS Glue, Amazon EMR and Apache Spark are named examples. Revision exercise: Implement a small batch transformation in Python and SQL, then describe how you would adapt its ingestion and recovery for streaming.
Automate transfer and backup
Posting: The role builds data-transfer and backup automation and supports data cataloging and discovery. Revision exercise: Write a small Python transfer script with a retry and verification step, then document the metadata needed to find and understand its output.
Review an IAM policy design
Posting: Cross-account roles, service control policies and permission boundaries are explicit responsibilities. Revision exercise: Draft a least-privilege cross-account access scenario, identify an overly broad permission and revise the policy design.
Rehearse infrastructure deployment
Posting: Hands-on infrastructure as code, containers, orchestration and managed cloud services are required; Terraform, Docker and Kubernetes are examples. Revision exercise: Review or build a small IaC configuration and container deployment, then explain how you would check compute, storage and networking dependencies.
Query and present platform costs
Posting: The engineer develops analytics and dashboards for platform usage and cost, with Amazon Athena and QuickSight named as examples. Revision exercise: Write SQL that aggregates sample usage and cost by account and workload, then sketch a dashboard view a stakeholder could use.
Use AI assistance with verification
Posting: Demonstrated use of tools such as Claude, Copilot or Cursor for code generation, review, debugging and documentation is required, alongside internal AI-assisted operational tools. Revision exercise: Use one named tool to review a small pipeline script, verify its suggestions with tests, and note a task where its output would need human scrutiny.

Your preparation checklist

Build a requirements evidence sheet
List factual evidence for the stated degree or equivalent, production data infrastructure experience, Python and SQL work, and collaboration. Mark gaps rather than inventing examples.
Rehearse a production-system walkthrough
Prepare one project diagram and a short account of your ownership, scale, reliability decisions, incidents, and measured results. The posting emphasizes operating production systems at scale.
Sketch an AI data architecture
As interview practice, draw how storage, a data lake, distributed processing, and vector retrieval could support LLM training, fine-tuning, evaluation, and RAG. Explain stack choices, transfer and backup automation, and data discovery.
Prepare a pipeline design exercise
Outline batch and streaming flows, orchestration, failure recovery, and a small Python or SQL example. Be ready to explain relevant experience with tools such as Glue, EMR, Spark, or Airflow without implying all are required.
Work through a multi-account access scenario
Draft a least-privilege access model using AWS Organizations concepts named in the posting: service control policies, cross-account roles, and permission boundaries. Explain how you would review access patterns for performance and cost.
Show cloud operations and cost evidence
Prepare an example of infrastructure as code, containers or orchestration, and managed cloud services you have operated. Separately, sketch usage and cost measures that an Athena and QuickSight dashboard could show stakeholders.
Explain an AI-assisted engineering example
Describe a real use of an AI tool for coding, review, debugging, or documentation, including how you checked its output. Propose a modest internal tool that could reduce manual platform operations; label it as an idea, not an existing AISG project.
Write questions about platform adoption
Bring questions on current data workloads, the boundaries of platform ownership, self-service needs for engineers and researchers, and how usage and cost visibility informs decisions. These follow the posting’s emphasis on usability and adoption.

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…