New opportunity

Senior Machine Learning Engineer

Viridien · Singapore

About this role

Viridien is seeking a Senior Machine Learning Engineer in Singapore to develop production-grade AI systems across the lifecycle from data ingestion to deployment. The role is part of a global AI team applying large language models and agentic workflows to digital solutions. The position uses a flexible hybrid work model.

Responsibilities

Deploy and fine-tune LLMs for domain-specific tasks. Build evaluation infrastructure and benchmarking pipelines to compare performance across datasets and tasks.
Apply reinforcement learning, knowledge representation and automated decision-making to complex problems.
Work with data scientists to build and own pipelines for data ingestion, transformation, versioning and labeling, maintaining reproducible data loops for iterative development.
Orchestrate multiple models and design agentic workflows using function calling, tools, RAG and code interpreters. Partner with production teams and software engineers to integrate AI solutions into products.
Provide technical leadership and mentor junior engineers.

Qualifications

A master’s degree or Ph.D. in Computer Science, Mathematics, Physics, Electrical Engineering or a related technical discipline.
Strong programming skills in at least one of C, C++, R, Java or Python; advanced understanding of LLM and VLM architectures, post-training methods including fine-tuning and reinforcement learning, and evaluation pipelines.
Experience in data and system engineering for large-scale model workflows, deep learning frameworks including TensorFlow or PyTorch, and enterprise machine learning and AI projects. Fluency with at least one of Cursor, Windsurf, Trae, Gemini CLI, Claude Code, Codex or a similar tool.
Ability to manage project deadlines and workloads, communicate progress to stakeholders, solve analytical problems and adapt to changing needs.

Preferred

Experience applying, fine-tuning, optimizing and deploying LLMs for industrial-scale real-user use cases is a plus.

Benefits include a 13-month annual base salary and performance bonus, personal and family healthcare, an onsite wellness and recreation corner, employee activities and family leave.

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

Skills for this role

Large Language Models (LLMs)Natural Language Processing (NLP)Vision-Language Models (VLMs)Model fine-tuningReinforcement learningModel evaluationBenchmarkingData engineeringData pipelinesData versioningData labelingAI agentsFunction callingTool useRetrieval-Augmented Generation (RAG)Knowledge representationDeep learningTensorFlowPyTorchCC++RJavaPythonCursorWindsurfTraeGemini CLIClaude CodeCodex

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

Domain-specific LLM development and evaluation
The role calls for fine-tuning and deploying LLMs, then building benchmarks across datasets and tasks. Prepare to explain how you would compare model versions and decide whether an improvement is meaningful.
Reproducible data infrastructure
The engineer will help own ingestion, transformation, versioning and labeling pipelines with data scientists. Prepare to describe how a data change can be traced through an iterative model-development loop.
Agentic workflows in production
The posting specifies multi-model workflows using function calling, tools, RAG and code interpreters, integrated with production teams and software engineers. Prepare to explain the boundaries between models, tools and product code.
Methods for complex decisions
Responsibilities include applying reinforcement learning, knowledge representation and automated decision-making. Prepare to discuss when each approach is useful for a problem, rather than treating them as interchangeable.
Technical leadership and delivery
The role includes mentoring junior engineers, managing deadlines across projects and keeping stakeholders informed. Prepare to describe how technical choices, workload and progress would be communicated.

Likely challenges to think through

Measuring domain performance without overfitting
A likely difficulty is improving a domain-specific LLM while preserving credible comparisons across multiple datasets and tasks. Prepare a framework for separating development data from evaluation data and investigating uneven results; this is a preparation hypothesis, not a reported Viridien problem.
Balancing agent capability with reliability
Orchestrating models and tools may improve task coverage while introducing more failure points during product integration. Prepare to explain how you would test individual steps and end-to-end behavior before deployment; the posting does not specify a particular reliability issue.
Keeping iteration reproducible under deadlines
Data ingestion, labeling and model changes can make fast experiments difficult to reproduce, while the role also requires balancing project workloads. Prepare to describe how you would prioritize changes and retain enough version history to explain results; this is an inferred trade-off.

Practice your answers

Technical practice

How would you choose between an LLM and a VLM for a domain-specific task?

What to coverThe posting calls for advanced understanding of both architectures. Suggested preparation: explain how input types, model capabilities, data availability and evaluation results would guide the choice.

How would you fine-tune an LLM and decide whether the result is ready to deploy?

What to coverFine-tuning and deploying LLMs for domain-specific tasks are stated responsibilities. Suggested preparation: discuss training data, a baseline, evaluation before and after fine-tuning, and trade-offs between performance and serving cost.

How would you build a benchmark that compares models across datasets and tasks?

What to coverThe role includes building evaluation infrastructure and benchmarking pipelines. Suggested preparation: cover representative datasets, consistent metrics and test conditions, error analysis, and how to avoid drawing conclusions from a single score.

Where could reinforcement learning and knowledge representation help an automated decision-making system?

What to coverThe posting names all three approaches as part of solving complex problems. Suggested preparation: describe what each contributes, how you would evaluate the resulting decisions, and when the added complexity may not be justified.

How would you design a pipeline for ingesting, transforming, versioning and labeling model data?

What to coverThe role involves owning these pipelines with data scientists. Suggested preparation: explain how you would track data changes, check transformation and label quality, and make a dataset usable for later model comparisons.

How would you make an iterative model-development loop reproducible?

What to coverThe posting explicitly calls for robust, reproducible data loops. Suggested preparation: describe how to connect dataset versions, code, model configurations and evaluation results so a change in performance can be investigated.

How would you orchestrate multiple models and tools in an agentic workflow?

What to coverThe stated work includes multi-model workflows with function calling and tool usage. Suggested preparation: outline how you would assign responsibilities, validate tool inputs and outputs, and handle failures without obscuring which component caused them.

How would you evaluate whether RAG or a code interpreter improves an agentic workflow?

What to coverBoth RAG and code interpreters appear in the workflow responsibilities. Suggested preparation: compare each addition with a simpler baseline, using task-relevant quality measures and considering latency, failure modes and operational complexity.

How would you troubleshoot a model workflow that performs well in development but poorly in a product?

What to coverThe role spans deployment and integration with production teams and software engineers. Suggested preparation: trace differences in data, model versions, orchestration and runtime behavior, then explain how you would isolate and verify a fix.

How would you use a deep learning framework and an AI coding tool without losing confidence in the implementation?

What to coverThe qualifications name TensorFlow or PyTorch and fluency with an AI coding tool or similar tool. Suggested preparation: discuss how you would test model code, review generated changes and check that an implementation remains reproducible.

Behavioural practice

Tell me about a time you mentored an engineer through a difficult technical problem.

What to coverTechnical leadership and mentoring junior engineers are stated responsibilities. Suggested preparation: use a situation, your actions and the result to show how you supported learning while maintaining technical standards.

Describe a time you worked with data scientists and software or production engineers to deliver an AI system.

What to coverThe role calls for collaboration across those groups. Suggested preparation: give a real situation, explain how you resolved differing needs or handoffs, and state the result.

Tell me about a time you balanced deadlines across competing projects.

What to coverThe posting calls for defining and managing deadlines and workloads across projects. Suggested preparation: describe the situation, your prioritization decisions and their result, including any trade-off you communicated.

Describe a time you had to keep stakeholders informed while an ML project encountered uncertainty.

What to coverRegular communication about progress is a stated qualification. Suggested preparation: explain the situation, what you communicated and when, and how your actions affected the outcome.

Tell me about a time an AI project changed direction after work was underway.

What to coverThe posting values flexibility when needs or direction change. Suggested preparation: give a real example covering the change, how you revised your approach and the result.

Sharpen your knowledge

Topics to prepare

LLM and VLM architectures
The posting requires advanced understanding of both. As a revision exercise, compare how an LLM and a VLM represent inputs, where their capabilities differ, and how those differences affect evaluation.
Fine-tuning and reinforcement learning
The posting names fine-tuning and reinforcement learning as post-training methods. Revise by outlining when each would suit a domain-specific task, what feedback it needs, and how you would check whether it improved the model.
Evaluation and benchmarking design
The role builds pipelines that compare models across datasets and tasks. Sketch an evaluation plan with datasets, metrics, baselines and repeatable runs; explain what each comparison would establish.
Reproducible data loops
The role owns ingestion, transformation, versioning and labelling pipelines. Trace a hypothetical dataset from ingestion to a model run, identifying what must be recorded to reproduce a result after data or labels change.
Agentic workflows and decision-making
The posting specifies multi-model orchestration, function calling, tools, RAG, code interpreters and knowledge representation. Draw a workflow showing which component makes each decision and how its output reaches the next component.
Production integration and technical leadership
The role integrates AI into products with production teams and software engineers, and mentors junior engineers. Rehearse a design review: explain a proposed system’s reliability trade-offs, ownership boundaries and a decision you would delegate or coach.

Skills to revise

Fine-tune and run a model
The role calls for domain-specific LLM fine-tuning and deployment, plus TensorFlow or PyTorch experience. As practice, use a small sample task to run a fine-tuning experiment in one of those frameworks, then demonstrate inference and record the configuration and results.
Build a repeatable benchmark
The posting explicitly calls for evaluation infrastructure. As practice, write a small benchmark in a listed programming language that runs two model variants on the same labelled examples and produces a per-task comparison.
Implement a versioned data pipeline
The role includes data ingestion, transformation, versioning and labelling. As practice, process a sample dataset through those stages and show how a particular model run can be traced back to its input and label versions.
Prototype an agentic flow
The posting names function calling, tool use, RAG and code interpreters. As practice, implement a small multi-step flow using a subset of those components and inspect its intermediate outputs; using all of them in one prototype is a preparation suggestion, not a stated requirement.
Code with an AI development tool
The posting requires fluency with at least one of Cursor, Windsurf, Trae, Gemini CLI, Claude Code, Codex or a similar tool. As practice, use one to help revise a small piece of model-pipeline code, then manually test and explain the changes.
Communicate delivery decisions
The role requires managing deadlines and workloads, keeping stakeholders informed, and mentoring junior engineers. As practice, draft a brief status update for a hypothetical AI integration: state progress, a technical risk, the next decision and who owns it.

Your preparation checklist

Assemble a requirements evidence sheet
List verifiable evidence for the stated degree, a listed programming language, TensorFlow or PyTorch, large-scale data and system engineering, enterprise AI work, and fluency with an AI coding tool. Mark gaps rather than inventing examples.
Prepare an end-to-end project walkthrough
Select a real project, if available, and outline the problem, your decisions from data ingestion through deployment, reliability measures, results, and what you would change. Industrial-scale LLM deployment is listed as a plus, not a requirement.
Build a model evaluation walkthrough
Prepare a sample benchmark plan for a domain-specific LLM task: datasets, baselines, metrics, comparison across tasks, error analysis, and how results would inform fine-tuning. Treat the plan as interview practice, not a stated company benchmark.
Diagram a reproducible data loop
Sketch ingestion, transformation, labeling, versioning, and validation, showing how a model run could be traced back to its data and repeated. The role explicitly calls for owning these pipelines with data scientists.
Practice an agentic-system design
Draw a hypothetical multi-model workflow using RAG, function calling, tools, or a code interpreter. Explain orchestration, failure handling, evaluation, and integration with production software engineers without assuming a specific Viridien product.
Review model and decision-making tradeoffs
Prepare concise explanations of LLM and VLM architectures, fine-tuning versus reinforcement learning, and where knowledge representation or automated decisions might help. Identify limitations rather than claiming every method fits every task.
Document leadership and delivery examples
Outline real examples, if available, of mentoring, balancing deadlines, adapting scope, and communicating progress to stakeholders. State your own contribution and the outcome for each.
Draft questions for the employer
Ask which domain-specific tasks and datasets matter first, how model quality is evaluated, who owns production integration, and what technical leadership looks like on the global AI team.

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…