New opportunity

Senior Software Engineer, Machine Learning Platform

Airwallex · Singapore

About this role

Role Purpose: You will build and scale the machine learning platform that powers risk decisioning at Airwallex, enabling teams to develop, deploy, monitor, and improve models that protect financial transactions. You will design reliable data and model infrastructure, productionize machine learning workflows, and enhance the speed and quality of experimentation across the Risk Platform. Responsibilities: Design and maintain robust model infrastructure; productionize ML workflows; collaborate with ML engineers, data scientists, product managers, and risk specialists to solve complex fraud and risk problems. Requirements: 5+ years of software engineering experience, with at least 3+ years focused on model training infrastructure, model serving systems, or MLOps platforms. Bachelor's or Master's degree in Computer Science, Engineering, or a related field. Hands-on experience with deep learning frameworks (PyTorch, TensorFlow, JAX) and training engines. Proficiency in Python, Java, or C++. Experience with distributed orchestration tools (Kubernetes, Ray, Kubeflow Pipelines, Airflow). Solid understanding of GPU architecture and hardware acceleration. Preferred Qualifications: Experience with model acceleration frameworks and Large Language Models (LLMs). Proficiency in performance profiling using tools like NVIDIA Nsight Systems. Experience with cloud platforms (AWS, GCP) and large-scale, low-latency production ML infrastructure. Working Arrangement: This role is based in Singapore and requires working from the office five days a week.

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

Skills for this role

PythonJavaC++PyTorchTensorFlowJAXKubernetesRayKubeflow PipelinesAirflowGPU architectureModel training infrastructureModel serving systemsMLOpsAWSGCPNVIDIA Nsight Systems

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

Build and Scale ML Infrastructure
The core responsibility is to design and maintain robust model training and serving infrastructure. You must demonstrate the ability to build systems that handle high-volume, low-latency financial transactions while ensuring reliability and scalability for the Risk Platform.
Productionize ML Workflows
You are expected to bridge the gap between experimentation and production. This requires hands-on experience with orchestration tools like Kubernetes, Ray, and Kubeflow to automate and stabilize the end-to-end machine learning lifecycle.
Optimize Performance and Hardware Utilization
Given the focus on GPU architecture and hardware acceleration, you must prioritize optimizing model training and inference performance. This involves using profiling tools like NVIDIA Nsight Systems to identify bottlenecks and improve efficiency.
Cross-Functional Collaboration
You will work closely with data scientists, risk specialists, and product managers to translate complex fraud problems into technical solutions. Success depends on your ability to communicate technical constraints and align infrastructure capabilities with business risk requirements.

Likely challenges to think through

Balancing Latency and Accuracy in Fraud Detection
The Risk Platform requires millisecond-level decisioning. A significant challenge is optimizing models for speed without sacrificing the accuracy required to prevent financial loss. Prepare to discuss trade-offs between model complexity, inference latency, and hardware acceleration strategies.
Navigating Ambiguity in Rapidly Evolving Fraud Landscapes
Fraud tactics evolve quickly, and the role requires owning the roadmap in an ambiguous environment. You may face situations where you must define technical requirements from scratch. Prepare by demonstrating how you apply first-principles thinking to solve undefined problems.
Managing Large-Scale Distributed Systems
Processing billions of events daily requires deep expertise in distributed orchestration. The challenge lies in maintaining system stability under heavy load. Be prepared to discuss your experience with scaling infrastructure and handling failures in distributed environments like Kubernetes or Ray.

Practice your answers

Technical practice

Designing Low-Latency Model Serving Infrastructure

What to coverDiscuss the architectural trade-offs when building a low-latency model serving system for high-volume financial transactions. Focus on strategies for minimizing inference time, such as model quantization, caching mechanisms, and efficient request batching.

Optimizing GPU Utilization for ML Workflows

What to coverExplain how you would profile and optimize GPU-based training or inference pipelines. Mention specific tools like NVIDIA Nsight Systems and how you identify bottlenecks in data transfer, kernel execution, or memory bandwidth.

Orchestrating Distributed ML Pipelines

What to coverDescribe your experience with tools like Kubernetes, Ray, or Kubeflow Pipelines for managing complex ML workflows. Focus on how you handle resource scheduling, fault tolerance, and scaling in a production environment.

Productionizing Deep Learning Models

What to coverDetail the lifecycle of moving a model from experimentation to production. Address versioning, automated testing, monitoring for model drift, and ensuring reproducibility across different environments.

Handling Data at Scale for Risk Decisioning

What to coverDiscuss the challenges of building infrastructure that processes billions of events daily. Focus on how you ensure data consistency and reliability when feeding streaming data into ML models.

Model Acceleration Frameworks

What to coverCompare different model acceleration frameworks and explain when you would choose one over another for production deployment. Discuss the impact on model accuracy versus inference speed.

Infrastructure for LLM Agents

What to coverGiven the role involves LLM agents for alert triage, discuss the unique infrastructure requirements for serving LLMs compared to traditional predictive models. Address memory management, latency, and cost considerations.

Hardware Acceleration Trade-offs

What to coverExplain the decision-making process for choosing between different hardware accelerators for specific ML tasks. Consider factors like power consumption, cost, and performance gains for training versus inference.

Behavioural practice

Owning a Technical Roadmap

What to coverDescribe a time you were responsible for defining and executing a technical roadmap without explicit instructions. Focus on how you identified the problem, prioritized tasks, and communicated progress to stakeholders.

Navigating Ambiguity in System Design

What to coverProvide an example of a project where you were given a high-level problem rather than a specific technical prescription. Explain how you broke down the ambiguity into actionable engineering steps.

Cross-Functional Collaboration on Complex Problems

What to coverDescribe a situation where you had to collaborate with non-technical stakeholders, such as risk specialists or product managers, to solve a technical challenge. Explain how you translated complex ML constraints into business-relevant outcomes.

Taking Initiative to Resolve System Failures

What to coverShare an experience where you identified a critical flaw or inefficiency in a production system and took ownership to fix it. Emphasize your decision-making process and the impact of your intervention.

Working in a Global, Distributed Team

What to coverDiscuss your experience working across different time zones and cultures. Provide an example of how you maintained effective communication and collaboration despite geographical distance.

Sharpen your knowledge

Topics to prepare

MLOps Lifecycle and Infrastructure
Review the end-to-end lifecycle of machine learning models, specifically focusing on productionizing workflows. Exercise: Map out a standard MLOps pipeline from data ingestion to model serving, identifying potential failure points in high-throughput, low-latency environments. This is relevant to the role's core responsibility of building and scaling the ML platform for risk decisioning.
GPU Architecture and Hardware Acceleration
Study the fundamentals of GPU architecture, including memory hierarchy, CUDA cores, and hardware acceleration techniques. Exercise: Explain how specific GPU optimizations improve training throughput or inference latency for deep learning models. This is directly requested as a requirement for optimizing model performance.
Distributed Orchestration Systems
Understand the architecture and use cases for distributed tools like Kubernetes, Ray, and Kubeflow. Exercise: Compare the strengths of Ray versus Kubernetes for managing distributed ML training jobs. This is essential for the role's requirement to manage large-scale, distributed ML infrastructure.

Skills to revise

Performance Profiling with NVIDIA Nsight
Gain proficiency in using NVIDIA Nsight Systems to identify bottlenecks in model training or inference. Exercise: Create a profile of a simple PyTorch model to identify GPU utilization gaps or data transfer bottlenecks. This is a preferred qualification for optimizing production ML systems.
Distributed Workflow Management
Develop hands-on experience with Kubeflow Pipelines or Airflow for automating ML workflows. Exercise: Design a DAG (Directed Acyclic Graph) that automates a model retraining loop, including data validation and deployment steps. This is critical for the role's focus on productionizing ML workflows.
Deep Learning Framework Implementation
Practice implementing and optimizing models using PyTorch or JAX. Exercise: Implement a custom training loop in PyTorch that utilizes distributed data parallel (DDP) strategies. This aligns with the requirement for hands-on experience with modern deep learning frameworks.

Your preparation checklist

Map MLOps experience to high-scale infrastructure
Prepare to discuss your 3+ years of experience in model training or serving systems. Create a summary of a past project where you optimized a machine learning workflow, specifically highlighting how you handled distributed orchestration (e.g., Kubernetes, Ray, or Airflow) to improve reliability or speed.
Prepare a deep dive on GPU-accelerated infrastructure
Review your experience with GPU architecture and hardware acceleration. Be ready to explain how you have optimized training or inference performance, potentially using tools like NVIDIA Nsight Systems to identify and resolve bottlenecks in large-scale production environments.
Develop a framework for adversarial systems design
Since the role involves protecting financial transactions, prepare to discuss how you design ML infrastructure that balances low-latency requirements with high accuracy. Consider how you would build systems that are resilient against evolving fraud patterns and how you would monitor model performance in real-time.
Formulate questions about the Risk Platform's scale
Prepare thoughtful questions for the interviewers regarding their current challenges with streaming pipelines or LLM agent integration. Ask about how the team balances the two-sided cost of fraud detection—minimizing false positives while preventing financial loss—to demonstrate your understanding of the business impact.
Reflect on your approach to autonomous problem-solving
Airwallex values 'founder-level energy' and ownership. Prepare specific examples of times you identified a technical gap, took initiative to fix it without being asked, and owned the outcome from design through to production deployment.

Availability last checked: 18 Oct 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…