I write software because the problems I actually wanted to solve turned out to be software problems. Fourteen years in production taught me exactly where creative workflows break, which is useful when you're building tools for creative people.

Now

ZygApp — a creator marketplace with affiliate product cards, member profiles and a Synergy Score that measures contribution. Django and PostgreSQL on Google Cloud Run, with a companion Chrome extension. I built it end to end, from schema to deploy pipeline. Full breakdown →

The market page on this site runs a stripped-down port of the same affiliate mechanics — ASIN-to-URL generation and a logged redirect, backed by a JSON file instead of Postgres. Same pattern, a tenth of the machinery, because four products don't justify a database.

Before that

Alter Concepts Gallery — my first web app. An interactive 3D gallery for previewing animated art with synchronised music playback, each piece linked to its minted token. Three.js, vanilla JS.

A run of CS50AI projects: cleaning data and training models to predict shopping-cart completion, image recognition with a TensorFlow neural network, a sentence parser extracting noun phrases, and a question-answering system using tf-idf ranking over a text corpus.

This site is Flask, Jinja and hand-written CSS — vanilla HTML, CSS and JS with no framework and no build step, which is a deliberate constraint rather than a limitation.

How I work

Reinforcement learning isn't only a concept in machine learning to me; it's roughly my method. Ship, look honestly at what happened, adjust, repeat. That's how the craft got good and it's how the code is getting there.