Machine Learning Engineer

About The Role

The role drives the development, scaling, and maintenance of core machine learning pipelines and real-time model serving infrastructure. This engineer will translate advanced prototypes into reliable, highly performant production systems that serve millions of predictions daily. Working within a collaborative team of data scientists and backend engineers, this role directly impacts system latency, model accuracy, and overall engineering velocity by building robust MLOps patterns.

Key Responsibilities

• Design, implement, and maintain scalable ML training and inference pipelines using Python, PySpark, and orchestration tools like Airflow or Prefect
• Deploy deep learning and classical ML models to production environments using AWS SageMaker, Kubernetes, and Triton Inference Server
• Develop and maintain a centralized feature store to ensure consistency between offline training data and real-time online features
• Implement automated monitoring, logging, and alerting systems to track model drift, performance degradation, and system latency
• Optimize model latency and throughput through techniques such as quantization, compilation (TensorRT/ONNX), and distributed serving
• Collaborate on architectural decisions to integrate ML services seamlessly with microservices-based backend APIs

What We Are Looking For

• 3-6 years of professional experience as a Machine Learning Engineer or Software Engineer with a heavy focus on production ML systems
• Strong proficiency in Python and hands-on experience with ML/DL frameworks such as PyTorch, XGBoost, or TensorFlow
• Solid software engineering fundamentals, including experience with Docker, CI/CD pipelines, and writing unit/integration tests
• Experience with cloud infrastructure (specifically AWS or GCP) and container orchestration using Kubernetes or EKS
• .S. or M.S. in Computer Science, Data Science, or a related quantitative engineering field
• Bonus: Experience with Triton Inference Server, MLflow, Ray, or running instruction fine-tuning for LLMs