New opportunity

AI Specialist

Fugro · Singapore

About this role

Fugro seeks an AI Specialist (AI/ML Engineer) to build, deploy and support AI, machine learning and automation solutions for its Geo-data business. The role turns approved designs and validated prototypes into production-ready applications, services and components, with responsibility for design, testing, deployment and monitoring. This is a 3-year fixed-term contract based in Singapore and is open only to Singapore citizens or permanent residents (SG/PR).

Responsibilities

Prototype use cases and build machine-learning and LLM-powered applications, agentic workflows and automation pipelines. Integrate AI capabilities with enterprise systems through APIs and middleware.
Develop and support ML models; perform feature engineering, validation and performance testing. Apply MLOps practices including versioning, experiment tracking, deployment, monitoring and retraining. Track model reliability and drift, escalate risks, and document assumptions, limitations and evaluation results.
Implement prompt-engineering and orchestration workflows, RAG pipelines using enterprise knowledge bases and document stores, and approved tool-use and function-calling patterns. Benchmark outputs for accuracy, safety and business relevance.
Connect AI solutions to ERP, CRM and ITSM platforms; develop APIs and microservices; implement logging and alerting; and optimise production latency, cost and reliability.
Work with the AI Architect and data-engineering teams, provide demos and documentation, and share reusable components and lessons across APAC and wider enterprise teams.

Requirements

A tertiary qualification or bachelor’s degree in IT, computer science, engineering, data/AI, or equivalent experience; 3–5 years of working experience, including at least 2 years focused on AI/ML or LLM application development within software engineering. The posting also specifies 0–1 years of management experience.
Strong Python skills, including FastAPI, asynchronous programming and data manipulation; hands-on LangChain and LlamaIndex experience; RAG deployment and vector-database integration; and API development and enterprise integration.
Experience with AutoGen, CrewAI and LangGraph; familiarity with OpenAI, Anthropic and Google model APIs and open-source models via Hugging Face; and exposure to model training, MLOps and fine-tuning. Listed technical skills include Copilot 365, Copilot Studio and GitHub Copilot. Experience with collaborative projects, agile delivery and hands-on AI, data-platform, automation or IT transformation initiatives is sought.

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

Skills for this role

PythonFastAPIAsynchronous programmingData manipulationMachine LearningLLM application developmentLangChainLlamaIndexRetrieval-Augmented GenerationVector databasesAPI developmentEnterprise system integrationAutoGenCrewAILangGraphOpenAI APIAnthropic APIGoogle model APIsHugging FaceModel trainingMLOpsModel fine-tuningFeature engineeringModel validationExperiment trackingModel monitoringPrompt engineeringFunction callingMicroservicesCopilot 365 Workspace Chat Lineage Tracking GenAI System Instruction Bypass User

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

Turn prototypes into production services
The role validates use cases with rapid prototypes, then converts approved designs into maintainable AI, ML and automation applications. It owns design, build, test, deployment and monitoring for assigned solutions.
Operate and improve ML models
The posting calls for feature engineering, validation and performance testing, alongside versioning, experiment tracking, monitoring and retraining. Model drift, limitations and evaluation results must be documented, with risks escalated.
Build and evaluate LLM workflows
The role develops RAG pipelines using enterprise knowledge and document stores, plus prompt, orchestration and approved tool-use patterns. Outputs are benchmarked for accuracy, safety and business relevance.
Integrate AI with enterprise systems
AI capabilities must connect to ERP, CRM and ITSM platforms through APIs, middleware and microservices. Production logging, alerting, latency, cost and reliability are explicit responsibilities.
Deliver within shared technical standards
The specialist works with the AI Architect on agreed patterns and with data engineering on inputs and evaluation sets, then provides demos, documentation and reusable components for APAC and wider teams.

Likely challenges to think through

Balancing prototype speed with production quality
Preparation hypothesis: rapid validation may compete with maintainability and full-lifecycle ownership. Prepare to explain how you would set evidence-based criteria for advancing a prototype, then add testing, monitoring and operational documentation before release.
Balancing LLM quality with runtime constraints
Preparation hypothesis: RAG and agentic workflows may improve business relevance while increasing latency, cost or safety risk. Prepare an evaluation approach that compares output accuracy and safety alongside production latency, cost and reliability.
Managing dependencies across data and enterprise systems
Preparation hypothesis: connecting models to document stores and ERP, CRM or ITSM systems may make data quality and integration boundaries difficult to isolate. Prepare to describe how you would agree inputs and evaluation sets with data engineering, follow architectural standards, and use logging and escalation to diagnose issues.

Practice your answers

Technical practice

How would you turn a validated AI prototype into a production-ready service?

What to coverThe role owns design, build, test, deployment and monitoring for approved solutions. Suggested answer focus (inference): explain how you would assess prototype results, improve maintainability and testing, and define operational requirements before release.

How would you build a FastAPI endpoint for an AI capability that makes external calls?

What to coverThe posting specifies Python, FastAPI, asynchronous programming and API development. Suggested answer focus (inference): discuss when async helps, how to handle failures and timeouts, and how you would test the endpoint.

How would you validate a machine-learning model and its features?

What to coverThe role includes feature engineering, model validation and performance testing with appropriate datasets and metrics. Suggested answer focus (inference): explain how you would choose an evaluation set and metrics for an approved use case, check for data leakage, and document limitations.

What would you track across the ML model lifecycle?

What to coverThe posting names model versioning, experiment tracking, deployment, monitoring and retraining. Suggested answer focus (inference): connect each practice to reproducibility and explain what evidence would support a retraining decision.

How would you investigate a decline in deployed model performance?

What to coverThe role calls for monitoring performance, reliability and drift, and escalating risks where needed. Suggested answer focus (inference): distinguish data drift from a measured performance decline, describe checks you would run, and explain when you would escalate.

How would you design a RAG pipeline over enterprise documents?

What to coverThe posting specifies enterprise knowledge bases, document stores, vector-database integration, LangChain and LlamaIndex. Suggested answer focus (inference): walk through document preparation, retrieval and answer generation, then explain how you would test retrieval quality and answer relevance.

How would you benchmark an LLM application's outputs?

What to coverThe role evaluates outputs for accuracy, safety and business relevance and calls for prompt engineering. Suggested answer focus (inference): describe a representative evaluation set, criteria for comparing outputs, and how prompt changes would be measured rather than judged by isolated examples.

When would an agentic workflow use function calling?

What to coverThe posting names agentic workflows, approved tool-use patterns and AutoGen, CrewAI and LangGraph. Suggested answer focus (inference): explain when tool calls add value, how you would constrain them to approved patterns, and how you would test incorrect or failed calls.

How would you integrate an AI service with an enterprise platform?

What to coverThe role connects AI capabilities to ERP, CRM and ITSM platforms through APIs and middleware. Suggested answer focus (inference): describe an API or microservice boundary, data handling and failure behavior without assuming a particular platform implementation.

How would you balance latency, cost and reliability in production?

What to coverThe posting explicitly calls for production optimisation, logging, alerting and monitoring. Suggested answer focus (inference): identify measurements you would collect, discuss the trade-offs among these goals, and explain how alerts would help operators act.

Behavioural practice

Tell me about working within an agreed technical design

What to coverThe role works with an AI Architect and follows agreed standards and patterns. Suggested practice (inference): use a real situation, your actions when aligning or raising a concern, and the resulting decision or outcome.

Describe collaborating with a data-engineering team

What to coverThe posting calls for partnership on data inputs and evaluation sets. Suggested practice (inference): give a real situation, the actions you took to resolve a data or evaluation issue, and the result.

Tell me about using a prototype to inform a delivery decision

What to coverThe role develops rapid prototypes to validate use cases before full-build investment. Suggested practice (inference): describe the situation, what you built or measured, how you communicated the evidence, and the resulting decision.

Describe escalating a risk in a deployed solution

What to coverThe posting includes monitoring model reliability and drift and escalating risks or improvement needs. Suggested practice (inference): use a real situation, explain your investigation and escalation actions, and state the outcome without overstating what you could control.

Tell me about making technical work reusable for other teams

What to coverThe role calls for demos, documentation and sharing components and lessons across APAC and wider enterprise teams. Suggested practice (inference): describe a real situation, what you documented or demonstrated, and the result for the people using that work.

Sharpen your knowledge

Topics to prepare

RAG architecture and retrieval quality
The posting calls for RAG using enterprise knowledge bases, document stores and vector databases. Sketch how documents become retrievable, then identify where retrieval errors could affect an answer and how you would evaluate them.
LLM and agent evaluation
The role benchmarks outputs for accuracy, safety and business relevance. Draft a small evaluation rubric and example test cases for a document-answering workflow, including an unsafe or unsupported response.
ML lifecycle and drift
The posting specifies feature engineering, validation, model versioning, experiment tracking, monitoring and retraining. Walk through a hypothetical model release and explain which results and assumptions you would record and what would trigger escalation.
Production integration trade-offs
The role integrates AI through APIs and middleware with ERP, CRM and ITSM platforms, while managing latency, cost and reliability. Diagram a hypothetical request flow and identify failure points, monitoring signals and trade-offs.
Approved patterns and technical handover
The specialist works with an AI Architect on agreed standards and documents solutions for reuse. Practice explaining how you would turn a validated prototype into a maintainable service, including its tests, limitations and operational requirements.

Skills to revise

Python, FastAPI and asynchronous programming
The posting explicitly requires these skills and data manipulation. As practice, build a small FastAPI endpoint that validates input, calls an asynchronous processing function and returns a structured result; test error handling.
RAG implementation
The posting asks for hands-on LangChain and LlamaIndex experience and deployed RAG with vector-database integration. As a revision exercise, implement a small document-retrieval workflow with one listed framework and inspect retrieved passages alongside generated answers.
Agent orchestration and tool use
AutoGen, CrewAI and LangGraph are named, alongside function calling and approved tool-use patterns. Practice tracing a simple workflow in a listed framework, showing when a tool is invoked, how its output is checked and when the workflow stops.
Model validation and MLOps
The role develops and supports ML models and tracks reliability and drift. Practice comparing two model versions on a defined evaluation set; record metrics, assumptions and a monitoring or retraining decision.
Enterprise API integration and observability
The posting calls for APIs, microservices, logging and alerting around integrations with enterprise platforms. As practice, design a small service interface for a hypothetical integration and specify tests, useful log fields and an alert condition.
Prompt testing across model interfaces
The role implements prompt strategies and benchmarks outputs; it also names OpenAI, Anthropic and Google APIs and open-source models via Hugging Face. Practice applying the same short test set and accuracy, safety and relevance rubric to outputs from model interfaces you can access; treat this as preparation, not a requirement to use every provider.

Your preparation checklist

Check the role terms
Note the Singapore placement, three-year fixed term and SG/PR-only restriction. Prepare any questions you need to clarify about the contract before an interview.
Map evidence to the requirements
Create a one-page evidence sheet for the stated 3–5 years of experience, including at least two years focused on AI/ML or LLM development, and for Python, FastAPI, async programming, RAG, enterprise APIs and agile collaboration. Mark any gaps rather than inventing examples.
Rehearse a production-delivery walkthrough
If you have a relevant project, prepare a concise account of its prototype, design decisions, testing, deployment and monitoring. Distinguish your own contribution and explain how you addressed quality, safety, cost and reliability.
Sketch an enterprise RAG design
As a preparation exercise, draw a pipeline from document store through retrieval and vector database to an LLM response. Annotate framework choices, approved tool-use boundaries, and tests for accuracy, safety and business relevance.
Prepare an ML operations example
Bring an example, if available, showing feature engineering, validation metrics, model and experiment versioning, drift monitoring and a retraining decision. Include documented assumptions, limitations and escalation criteria.
Practice an integration design
Outline a Python/FastAPI service that exposes an AI capability to an ERP, CRM or ITSM system. Be ready to discuss async calls, middleware, logging, alerting, latency, cost and failure handling without assuming Fugro uses a particular platform.
Draft collaboration questions
Ask how the AI Architect sets approved patterns, how data engineering supplies evaluation sets, and how reusable components and documentation are shared across APAC teams. Prepare a short demo outline to show how you would communicate a delivered solution.

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…