Can Haskell Become a Great Language for Data Science? | Michael Chanvinda | ZuriHac 2026

Michael Chanvinda explores how Haskell can model DataFrames using advanced type-level programming to bridge the gap between interactive data science and reliable software.

Michael Chanvinda presents a comprehensive exploration of how Haskell can become a more effective language for data science, exploratory analysis, and data engineering. The talk focuses on reducing the gap between rapid experimentation and reliable software development by creating tools that combine the immediacy of interactive workflows with the safety and expressiveness of Haskell’s type system. A central theme is the role of DataFrames as the dominant abstraction for working with tabular data. Drawing on the history of DataFrames from S, R, pandas, and modern systems such as Polars, the presentation examines their underlying semantics, their relationship to schemas and types, and the challenges of representing flexible, real-world data in a strongly typed language. The talk discusses how Haskell can model DataFrames using advanced type-level programming while avoiding the usability issues that have limited adoption in previous approaches. The presentation also introduces the idea of “Immediate Haskell”, an approach that combines progressive typing with interactive data exploration. Through practical demonstrations, Michael shows how users can load datasets, inspect statistics, create visualizations, perform feature engineering, train machine learning models, and export results, all within a notebook environment that supports reactive computation. A significant portion of the talk explores reactive notebooks for Haskell, inspired by systems such as Jupyter, Pluto, and Marimo. Topics include reproducible analysis, dependency tracking, notebook architecture, Markdown-based notebooks, interoperability with other languages, and converting exploratory work into reusable Haskell modules. The session concludes with a discussion of data infrastructure, Apache Arrow, Parquet, lazy DataFrames, performance considerations, and the future of the Haskell data ecosystem. Chapters 00:00:00 Introduction And Motivation 00:05:14 Exploring Data With Interactive Notebooks 00:08:49 Why DataFrames Matter 00:12:33 Formalizing DataFrames And Types 00:15:43 Type-Level DataFrames In Haskell 00:18:33 Ergonomic Data Processing APIs 00:20:43 Performance And Apache Arrow 00:23:05 Building A Haskell Data Ecosystem 00:24:13 Literate Programming And Notebooks 00:25:50 Reactive Notebooks Explained 00:28:25 Notebook Architecture And Package Management 00:31:22 Notebook Demonstration And Reporting 00:38:53 Exporting Notebooks To Haskell Modules 00:42:43 DataFrames, Grouping And Query Execution 00:47:58 The Future Of Data Haskell 00:48:43 Q&A And Ecosystem Discussion About the Speaker Michael is a Staff Software Engineer at CoreWeave where he works on marimo notebooks. Michael is also a core contributor to the DataHaskell organisation working primarily on data infrastructure and explainable AI. Michael Chanvinda presented this talk at ZuriHac 2026, held at OST – Eastern Switzerland University of Applied Sciences in Rapperswil-Jona, Switzerland. Michael was i