Practical LLMs for Modern Data Science

Course details

Learn how large language models (LLMs) can be used as practical assistants in everyday data science work. Start by seeing LLM-generated code in action and then build your own reusable helpers that speed up repetitive tasks such as exploration, preprocessing, and boilerplate modeling. Find out how to use LLMs to generate code that can be inspected, executed, and validated, rather than trusting black-box outputs. Explore how LLMs behave in classification and regression tasks and how to evaluate those results using standard data science metrics.

Learning objectives
Build reusable LLM-powered helpers that generate inspectable Python code for common data science tasks.
Use LLMs effectively across key stages of the data science workflow, including exploration, preprocessing, modeling, and evaluation.
Apply LLMs to classification and regression tasks using prompt-based approaches and validate the results with standard metrics.
Decide when to rely on LLM assistance and when human judgment, validation, and domain knowledge are required.

Concepts

Introduction - Get Productive Fast

How LLMs fit into data science
Building your first LLM helper - From prompt to inspectable Python code

Exploratory Data Analysis Helpers

Generate an EDA summary helper
Extend the EDA helper with visual diagnostics
Validate and refine EDA helper outputs

Preprocessing Helpers

Build a data cleaning helper
Create a feature encoding and scaling helper
Combine preprocessing helpers into a reusable pipeline

Modeling Helpers

Build a baseline classification helper
Build a regression modeling helper
Refine and customize model helpers

Evaluation Helpers

Build a classification evaluation helper
Build a regression evaluation helper
Validate model performance with baselines

Judgment, Control, and Responsible Use

Make LLM helpers auditable and reproducible
Run helpers with local or privacy-preserving LLMs

End-to-End Application

Run the full helper-based workflow
Stress-testing helpers on a new dataset

Wrap-Up and Next Steps

Patterns for building your own LLM helpers