ML Engineer

Responsibilities:
Design, train, and fine-tune deep learning models in PyTorch across computer vision (image classification, segmentation, and detection) and time series domains (rPPG signal processing for heart rate, SpO2 and respiratory rate estimation).
Own the full ML lifecycle for assigned models from dataset curation and preprocessing through to training, evaluation, deployment, and post-launch monitoring on Azure ML managed online endpoints.
Use MLflow (and Azure ML's integrated experiment tracking) to log experiments, version artifacts, register models, and drive a reproducible, auditable model development workflow.
Build and maintain training, evaluation, and inference pipelines in Python, integrating with Azure Blob Storage for datasets, Azure Container Registry for scoring images, and the gated Test Prod ML Registry promotion workflow.
Optimize model inference latency, accuracy, and cost across CPU and GPU endpoints, ensuring vitals and skin assessment endpoints meet real-time UX requirements on mobile clients.
Implement bias testing, robustness evaluation, and clinical validation protocols across demographic groups, particularly for skin-tone fairness in dermatology models and age/sex coverage in vitals models.
Collaborate with backend engineers to expose models via REST/WebSocket inference contracts consumed by Azure Functions and with mobile engineers on input-quality requirements from camera capture.
Set up drift, latency, and quality monitoring across deployed endpoints using Azure Monitor, Application Insights, and New Relic, and respond to inference quality regressions.
Leverage AI agents and AI coding assistants daily for literature review, code generation, dataset exploration, experiment design, and accelerating training iteration cycles.
Contribute to ML platform improvements, closing gaps in model approval workflows, drift monitoring, A/B testing, and CI/CD for model promotion.
Requirements:
Strong Python engineering skills with production-grade discipline in typing, testing, packaging, and code review.
Deep, hands-on PyTorch experience training and fine-tuning modern deep learning architectures (CNNs, transformers, U-Net / segmentation models, and sequence models).
Solid grounding in computer vision: image classification, segmentation, detection, augmentation strategies, and medical or biometric image handling.
Time series modelling experience, ideally in signal processing, physiological signal estimation, or rPPG/video-derived vitals; comfort with sequence models, filtering, and temporal feature engineering.
Proficiency with MLflow for experiment tracking, model registry, and reproducible runs; experience with Azure Machine Learning is a strong advantage.
Experience deploying models as low-latency inference services (managed endpoints, containerized scoring, REST APIs) and integrating them into a wider cloud platform.
Familiarity with Azure ecosystem services in an ML context (Blob Storage, Container Registry, Key Vault, Functions) or a clear track record on AWS/GCP that translates.
Strong AI tool fluency, comfortable using LLM-based coding assistants, AI-augmented research and dataset workflows, and AI-driven debugging in day-to-day work.
Leadership and Communication:
Able to communicate model design decisions, trade-offs, and limitations clearly to engineering, product, and clinical stakeholders.
Bias toward ownership and shipping; comfortable taking a model from a notebook to a monitored, versioned, production endpoint.
Collaborative across disciplines, particularly with clinical or domain experts, mobile engineers consuming inference outputs, and platform engineers operating shared infrastructure.
Calm and rigorous under pressure when production model issues arise, including quality regressions on PHI-handling endpoints.

Experience
3-7 yrs