Lead Enterprise Data Architect at Cytel, Richmond, TX

Title of the Talk:
Building Reliable AI Pipelines: From Data Quality to Scalable LLM Evaluation with Power BI and SQL

Abstract of the Talk:
AI models are only as good as the data that fuels them. Yet, many organizations still struggle to implement end-to-end pipelines that are reliable, transparent, and scalable enough to meet modern AI demands—especially in production environments. This keynote addresses the growing need for robust AI operationalization by focusing on how to embed trust, accuracy, and accountability into the very foundation of AI pipelines.

Drawing on real-world enterprise use cases and more than 15 years of implementation experience, Milan Parikh outlines a blueprint for designing intelligent pipelines using SQL-powered data workflows, ML model validation frameworks, and Power BI-driven observability. The session explores how to move beyond proof-of-concept AI projects toward sustainable, production-grade pipelines that incorporate repeatable data quality checks, robust metadata lineage, and scalable evaluation strategies for large language models (LLMs).

Attendees will gain insight into how governance, continuous monitoring, and feedback loops can be integrated seamlessly into AI lifecycle management—bridging the gap between data engineering, ML teams, and business stakeholders. The session also highlights how Microsoft’s data ecosystem—including Power BI and Synapse/Fabric—is being used to drive real-time analytics, streamline root cause analysis, and support automated data health assessments in AI applications.

Whether you’re an enterprise architect, data scientist, or decision-maker, you’ll walk away with proven strategies to build AI pipelines that scale confidently, adapt rapidly, and earn user trust—turning AI from a risk into a resilient asset..

.