My career has had two arcs. The first was independent consulting: I started professionally at sixteen, kept it up through college, and built a practice around long client relationships. The second was Loomly, where I spent nine years as founding engineer, growing a client project into a profitable SaaS company and scaling the product, infrastructure, and team through two acquisitions.
My parents brought home a 486 when I was eight, and I spent the next few years taking computers apart, building machines for friends, and poking at Linux. Around age thirteen I picked up a book on Perl and started programming; one of my first real projects was an online catalog for my dad's historic-newspaper business, a single Perl script backed by a CSV file. From there I worked my way into PHP and MySQL, and at sixteen I met my first paying client on the WebHostingTalk forums. Through high school I built web applications for paying clients and learned how to ship software that people depended on.
I studied Information Technology and Economics at RIT in Rochester, New York, though most of my engineering skills were self-taught, mostly from books. I found Ruby on Rails after years of writing PHP without much framework help, and it became the technical center of my work. Rochester's small Ruby community kept me in part-time consulting work through school.
After graduating I committed to consulting full time. A year later I moved to San Diego, got involved with SDRuby, and found steady work through the local Ruby scene. I usually had several clients at once, many for years at a time. Along the way I spent a year without a fixed home base, working from wherever I happened to be.
In January 2016 I got paired with a first-time founder at a local Ruby workshop night. He had an idea and a spreadsheet for managing social media posts. I took him on as a client, and that small engagement became the next nine years of my career.
That spreadsheet grew into Loomly, a social media management platform and a profitable company of twenty-five-plus people, acquired in 2021 and again in 2025. As founding engineer I stayed with it from early product work through years of Rails and infrastructure upgrades, scaling problems, team growth, and the handoff after the second acquisition. It was the longest and deepest work of my career, and I'm proud of the product and the team we built.
After the 2025 acquisition I left to take an extended learning sabbatical. I planned it as an 18-month break from normal client work: time to study broadly, from relearning long-forgotten math to digging into machine learning, systems programming, and new languages, and to build things as I went. It also left room for Mandarin study and jazz piano.
Language-learning platforms I built from scratch and run myself. They generate stories and sentences from the vocabulary a learner already knows, with audio in regional accents and spaced-repetition review. Built with TypeScript, they include background job processing, text-to-speech generation, LLM integration, and payments. I built Mandarin Forge primarily for my own HSK 3 study and still use it daily; Spanish Forge brings the same approach to Spanish.
A private, map-first archive for Foursquare and Swarm check-in history. Upload an account export to map and search every check-in, replay journeys, and view calendar heatmaps and photos. Profiles are private by default, with controls for what to share. Built with SvelteKit and Rust.
A jazz theory reference for chords, scales, and progressions in all twelve keys, each written out on the staff with common voicings, playback, and printable practice sheets. You can type in chord changes to get a roman-numeral analysis, or the notes you're holding to see which chords they spell. I first built it in 2010–2012 and rebuilt it in 2026 on Rails 8.
A rebuild of my longest-running client's Rails application (powering their entire business) onto Rails 8 and modern terraformed infrastructure, plus a full redesign of the storefront. Done largely pro bono, partly as a way to explore the current generation of coding agents.
An embedding model for jazz chord progressions: a small transformer trained with contrastive learning over ~6,900 standards, encoded so a tune and its transpositions map to the same vector. Searching by harmony rather than melody surfaces contrafacts like Countdown and Tune Up. The model and an interactive demo are on Hugging Face.
github.com/eigenben/jazz-harmony-embeddings → demo on Hugging Face →
A framework for generating programming exercises with LLMs, including LoRA fine-tuning of open models and serving with vLLM.