Reproducible Polyglot Data Science

This book is the culmination of my previous works and experiences, including:

• Reproducible Analytical Pipelines with R
• The {rix} R package for managing Nix environments
• The {rixpress} R package for orchestrating Nix-based pipelines
• The ryxpress Python package for orchestrating Nix-based pipelines

After writing my first book and starting on a Python edition, I realized that solving the reproducibility problem one language at a time was a flawed approach. The real solution needed to be universal. That universal solution is Nix.

Since discovering Nix, I built several tools on top of it: {rix}, {rixpress}, and ryxpress. They solved real problems, but they were tied to individual languages. Then came large language models (LLMs), which I believe will be a game changer: you, as the data scientist, can focus on the actual data science and leave the mechanical plumbing — environment setup and pipeline code — to LLMs. That insight is what led me to build T.

T is a new programming language that I designed and implemented with the help of LLMs. It is a domain-specific language specialised for pipelines: its only purpose is to make writing, inspecting, debugging, and running pipelines as simple as possible. T is designed to be used both by humans who want full control and by LLMs acting on their behalf. It coordinates R, Python, and Julia, and it is built on top of Nix: every T project is a Nix project, and Nix is used both to set up the reproducible environment and to run the pipeline. See the T documentation for more.

This book moves beyond language-specific tools. Instead, it shares a vision for what robust, polyglot data science should be, with Nix as the foundational layer and T as the language that makes it all possible.

What You'll Learn

• Master the Nix Philosophy: Understand why Nix is the ultimate tool for solving the "it works on my machine" problem once and for all.
• Build Bulletproof Environments: Use T to declaratively define exact software environments, pinning versions of R, Python, Julia, their packages, and all system dependencies.
• Orchestrate Polyglot Pipelines: Use T to build, run, and debug complex analytical pipelines that seamlessly pass data between different languages.
• Pair with LLMs: Treat LLMs as first-class collaborators, using T's structured feedback and built-in agent onboarding to let them write and verify the pipeline plumbing for you.
• Automate and Distribute: Package your pipelines with Docker and automate your entire workflow with GitHub Actions, all built on your reproducible Nix foundation.

As for all my books, Reproducible Polyglot Data Science will always stay free to read online:

https://reproducible-data-science.dev/