See LionPATH to schedule.
Course Topic: Agentic Data Science: Autonomous Workflows, Tool Use, and Human-in-the-Loop Systems
This course provides hands-on training in creating dependable agentic data science systems and analytics workflows that can use generative AI and integrated tools such as SQL, Python, APIs, and structured human oversight to plan, reason, and carry out end-to-end tasks. Students start by learning the fundamentals, such as what agentic systems are, how to prompt large language models properly for analytics, and how to safely and correctly invoke tools within multi-step workflows.
In addition to learning how to coordinate intricate analytical pipelines that include exploratory data analysis, data cleaning, feature engineering, modeling, and interpretation, students also learn how agents access and interpret real data using data dictionaries and metric definitions. Throughout, the focus is on creating systems that, despite automating increasingly complex analytical decisions, are transparent, auditable, and accurate.
Operational credibility is one of the course's main themes. Students build tools to check how accurate, consistent, and reliable their systems are on a large scale; set up safety measures to prevent issues such as data leaks, wrong prompts, and unexpected behavior; and use methods that involve human oversight.
Students build on their work in the second half of the course by testing how strong their systems are with fake and altered data, experimenting with agents to improve features and choose models, and using techniques to ensure their systems are ready for production, such as managing costs, creating operational guides, and ensuring students finish a midterm agentic EDA and data quality triage mini-build as part of the project-based assessment process. Students then go on to a capstone project in a group that produces a comprehensive end-to-end agentic analytics system with governance, monitoring, and an evaluation report based on evidence.
By the end of the course, students will have a thorough understanding of the operational risks as well as the technical capabilities of GenAI-enabled data workflows, enabling them to be responsibly designed, evaluated, and implemented in real organizational settings.
