Primary speaker:
Mari Motrich, Manager, Data and Analytics, University Registrar’s Office, Vice-Provost, Strategic Enrolment Management
Additional speakers:
Jason Lu, Senior Information Analyst, University Registrar’s Office, Vice-Provost, Strategic Enrolment Management
Joyce Dong, Senior Information Analyst, University Registrar’s Office, Vice-Provost, Strategic Enrolment Management
Titus Hsu, Azure Platform Engineer, Enterprise Apps & Solutions Integration, Office of the Chief Information Officer
Description:
This presentation outlines the outcome from a proof-of-concept project to migrate a strategic on-premises Machine Learning Yield Prediction solution and its associated VPSEM Admission Data Mart to the Microsoft Fabric platform and Azure Machine Learning. Led by the University Registrar Office Data and Analytics team, with support from the EASI Reporting and Analytics Technology team, this initiative modernizes a legacy analytics ecosystem to deliver scalable, governed and operationally efficient advanced analytics capabilities.
The session will walk through the end-to-end transformation — from infrastructure provisioning and Lakehouse architecture design to ETL/ELT pipeline reconstruction using Fabric Data Factory and Spark, and the refactoring and deployment of machine learning models within Azure ML’s MLOps framework. Attendees will gain insight into how model reproducibility, automated training and deployment, and monitoring pipelines were implemented to improve model accuracy while reducing technical debt.
We will also detail the transition from Tableau to Power BI, leveraging Direct Lake connectivity to improve performance and enable real-time analytics. Particular emphasis will be placed on strengthening data governance, implementing role-based access controls and mitigating migration risks such as data integrity, pipeline disruption and model degradation.
The presentation highlights practical lessons learned in dependency management, phased migration strategies, parallel pipeline runs and stakeholder knowledge transfer. By the end of the session, participants will understand how to architect and execute a large-scale migration of machine learning workloads to a unified analytics platform — balancing innovation, governance and operational continuity — while positioning institutional analytics for future growth.
