AI Lifecycle Management: MLOps in Practice

Day 3 — Model serving, Optimisation, and Automation for Production MLOps

The final day focuses on operationalising machine learning in production. Participants deploy an AI model as a containerised service on a server, expose it via an API, and connect the key building blocks around the service, such as monitoring and a lightweight front-end to create an end-to-end ML application. Next, we look at performance and scalability. You will optimise inference by exporting models to ONNX and running them with ONNX Runtime, and we discuss what professional inference serving looks like in GPU-enabled production environments using an inference server.

After lunch, a guest lecturer from Captic shares how they deploy AI-driven vision and robotics for industrial automation, with a strong focus on robust edge deployments and real-world data collection in demanding production settings, especially food-related environments.

In the afternoon, participants build an automation workflow with a data orchestration framework to make training, packaging and deployment repeatable. The day ends with an overview of ML cloud platforms and services (AWS SageMaker, Azure ML, Vertex AI) and how to position them in an MLOps architecture.

Lecturer: Alexander D'hoore (VIVES)

Industry case: Tim De Smet (Captic): Co-founder and CTO of Captic, an AI-vision and robotics company developing plug-and-play physical AI systems for industrial environments. Previously Squad Lead at ML6. Background in machine learning and computer vision, with hands-on experience in bringing AI from prototype to production in demanding, real-world settings.