Principal Engineer - Machine Learning

About the role

Western Digital is looking for a Principal Engineer - Machine Learning in Singapore. Unlike typical platform roles where work is absorbed by a large team, this person is the primary owner of ML systems that detect product development defects, model material behavior with limited data and select high-value experiments from an active learning pipeline. The models run in product development and their impact is immediate.

Key responsibilities

• Build, train, evaluate and maintain CNN/U-Net/ViT models for automated inspection and measurement, owning performance end-to-end including ablation studies, confidence calibration and product development monitoring.
• Build and maintain real-time anomaly detection for sensor and time-series data streams, including statistical baselines, threshold calibration and drift alerting.
• Own surrogate model pipelines and active learning systems, configuring acquisition functions and integrating with versioned feature sets.
• Own the data contract between the Data Engineer and the model stack: define feature specifications, validate datasets and escalate data quality issues early.
• Maintain model versions, training pipelines and containers, contribute to MLflow tracking and CI/CD, and monitor product development reliability.
• Mentor junior team members and document model design decisions and evaluation outcomes for production handoff.

Required profile

• Bachelor's or Master's degree in AI, Machine Learning, Computer Science or a related field; AI major or strong AI research focus preferred.
• 1-3 years of hands-on ML engineering experience, or equivalent depth shown through internships, academic research or open-source contributions.
• Demonstrated component-level ownership of an end-to-end ML pipeline (training through deployment).

Required skills

• Python and PyTorch (independent model training and evaluation)
• Computer vision: CNNs, and hands-on depth in U-Net/segmentation, ViT/transformers or time-series anomaly detection
• Surrogate modeling and active learning (acquisition functions, uncertainty-guided scheduling)
• Data-to-model interface: feature specs, dataset validation, data quality checks
• Practical MLOps: MLflow, Docker, Git, basic CI/CD
• Model evaluation and uncertainty analysis: ablation studies, confidence calibration, validation methodology
• Technical documentation to production-handoff standard

Nice to have: PINNs, Bayesian methods, reinforcement learning basics, AWS fundamentals (S3, EC2, SageMaker) and RAG pipeline fundamentals.