Sabbatical
Following the acquisition of Loomly in early 2025, I left to take an 18-month learning sabbatical. After nine years building one product, I wanted a stretch of time for deliberate study, especially machine learning and AI engineering.
I started with the math, relearning linear algebra and multivariable calculus. From there I moved through probability and statistics, classical machine learning, deep learning, transformers, and the practice of shipping AI-backed applications. One of the exercises was building a small LLM from scratch. I attended NeurIPS in San Diego in December 2025, and I spent some of the year on Rust systems programming too.
It wasn't all technical. I kept studying Mandarin Chinese (HSK 3) and jazz piano, two slower projects that fit the same rhythm: steady practice over a long period of time.
Built along the way
- MandarinForge, a Mandarin-learning platform, ~100k lines of TypeScript, built from scratch
- Rare Newspapers, a Rails 8 rebuild of my longest-running client's production app, partly to explore coding agent capabilities and workflows
- penguinoh_generator, a text-to-image generator, fine-tuning the FLUX.1-dev diffusion model with DreamBooth and LoRA
- python-exercises-generator, a small framework for generating programming exercises with LLMs, then testing fine-tuned open models
Books read
Since I was a teenager I've enjoyed learning through structured technical books, and I spent a lot of the sabbatical reading. This is the list, grouped by theme and pulled from my Goodreads.
Mathematics & statistics · 7
- No Bullshit Guide to Math and Physics ●●●●●Ivan Savov · Feb 2025
- No Bullshit Guide to Linear Algebra ●●●●●Ivan Savov · Mar 2025
- Essential Math for Data Science ●●●●●Hadrien Jean · Apr 2025
- Practical Statistics for Data Scientists ●●●●●Peter Bruce · Apr 2025
- Why Machines Learn ●●●●●Anil Ananthaswamy · Aug 2025
- Practical Linear Algebra for Data Science ●●●●●Mike X. Cohen · Aug 2025
- Mathematics of Machine Learning ●●●●●Tivadar Danka · Sep 2025
Machine learning & AI engineering · 13
- Hands-On Large Language Models ●●●●●Jay Alammar · Jul 2025
- Build a Large Language Model (From Scratch) ●●●●●Sebastian Raschka · Jul 2025
- Machine Learning with PyTorch and Scikit-Learn ●●●●●Sebastian Raschka · Sep 2025
- Machine Learning Q and AI ●●●●●Sebastian Raschka · Sep 2025
- Fundamentals of Deep Learning ●●●●●Nithin Buduma · Sep 2025
- AI Engineering ●●●●●Chip Huyen · Oct 2025
- Generative AI with Python and PyTorch ●●●●●Joseph Babcock · Nov 2025
- Hands-On Generative AI with Transformers and Diffusion Models ●●●●●Omar Sanseviero · Dec 2025
- Python Machine Learning By Example ●●●●●Yuxi (Hayden) Liu · Dec 2025
- Generative AI Design Patterns ●●●●●Valliappa Lakshmanan · Mar 2026
- Transformers in Action ●●●●●Nicole Koenigstein · Mar 2026
- LLMs in Production ●●●●●Christopher Brousseau · Apr 2026
- The Welch Labs Illustrated Guide to AI ●●●●●Welch Labs · Apr 2026
Python & data science · 3
- Python Data Science Handbook ●●●●●Jake VanderPlas · Aug 2025
- Modeling and Simulation in Python ●●●●●Allen B. Downey · Aug 2025
- Fluent Python ●●●●●Luciano Ramalho · Mar 2026
Systems programming · 3
- Rust for Rustaceans ●●●●●Jon Gjengset · Feb 2025
- Mastering Go ●●●●●Mihalis Tsoukalos · May 2026
- Asynchronous Programming in Rust ●●●●●Carl Fredrik Samson · Jun 2026
Computer science & algorithms · 2
- Grokking Algorithms ●●●●●Aditya Y. Bhargava · Jun 2026
- Data Structures the Fun Way ●●●●●Jeremy Kubica · Jul 2026
Software engineering & tooling · 4
- LazyVim for Ambitious Developers ●●●●●Dusty Phillips · Oct 2025
- Programming TypeScript ●●●●●Boris Cherny · Nov 2025
- The Software Engineer's Guidebook ●●●●●Gergely Orosz · Apr 2026
- Software Architecture: The Hard Parts ●●●●●Neal Ford · Jun 2026
Adjacent reading · 3
- Superagency unratedReid Hoffman · May 2025
- The Scaling Era: An Oral History of AI ●●●●●Dwarkesh Patel · Apr 2026
- The Marginal Revolution ●●●●●Tyler Cowen · Apr 2026
What's next
The sabbatical was planned as 18 months, and it's getting toward the end of that window. I'm starting to look for the next long-running engineering problem to work on. If you'd like to get in touch: me@benhughes.name.