Senior full-stack engineer. Currently obsessed with putting LLMs on a leash.
"What I cannot create, I do not understand." - Feynman
I've spent the last seven years shipping web products, mostly the Node.js / Nuxt / React kind that have to stay up while real users are mid-upload. These days I'm pivoting hard into AI: local-first coding agents, Claude Code plugins, MCP servers, and the occasional weekend spent breaking my own classifiers. This page is a snapshot of what I'm building and where my head is at.
I'm a Senior Software Engineer at TransMedia Inc. (Toronto, remote). Day-to-day I own freeconvert.com and chunks of the wider TransMedia platform, set up the Jenkins + Docker pipeline behind ProPDF, and shipped TrueRCE, an in-house rich-text editor we use as a drop-in TinyMCE replacement. I've been there since 2021 and got promoted into the senior seat in January 2025.
I keep a running list of things I'm actually building, not things I "have experience with." If a thing isn't on this list it's because I haven't shipped it lately.
| Repo | What it is |
|---|---|
| nip | A local-first CLI code agent on top of Gemma + Ollama. ReAct loop, tool-calling, no cloud calls. Reads, writes, refactors code on your machine. |
| claude-human-review | A Claude Code plugin that narrates every edit the agent makes in plain English. Born out of distrust of silent diffs. |
| article-writing-skills | Claude skills modelled on the writing voices of specific engineers and researchers. A testbed for how far prompt-shaped style transfer can go. |
| mern-mcp | An MCP server that scaffolds full-stack CRUD for MERN apps, Mongoose models through React components, with a preview-before-apply step so the agent stays on a leash. |
| bg-remover-ai | A serverless background remover that runs entirely in the browser via ONNX. No uploads, no tracking. An experiment in how much real ML you can ship to the client in 2026. |
| mars-snap | A shader-driven procedural Mars landscape generator with atmosphere, fog, and filmic lighting. Wallpaper-quality PNGs. Pure GLSL. |
A few smaller things I keep maintained: EnviroConfig (env config + schema validation for Node), node-express-rate-limiter (Redis-backed), vue3-counter, ProPassword, and Ayah Daily.
- Local-first coding agents, Ollama, Gemma, ReAct loops, the question of how small a model has to be before agency falls apart.
- Claude Code as a platform, not a product. Skills, plugins, hooks, MCP servers, a real surface to build on.
- Adversarial examples in NLP. I trained a spam classifier and then spent a weekend systematically wrecking it; turns out targeted unicode substitutions are still embarrassingly effective. (I suspect this is going to be a recurring theme.)
- Browser-native inference with ONNX Runtime. The client is more capable than most people are using it for.
Not "things I've touched", things I'd cheerfully open a fresh repo with on a Monday morning.
Languages
Frontend
Backend & Data
Infra & DevOps
AI / Agents
I write up things after I've actually done them. The ones I think held up:
- Breaking my own spam classifier, a hands-on tour of adversarial examples (Apr 2026)
- How Node.js handles 10,000 requests (Mar 2026)
- When your VPS gets hacked: how I found a miner running on our server (Dec 2025)
- Dive into V8: how JavaScript gets supercharged (and sometimes slows down) (Aug 2025)
- Turn a phrase into a strong password (and why it works) (Oct 2025)
The rest are on Medium, Node internals, V8, Docker, AWS, security incident reports, and a growing pile of AI deep-dives.
The thing I most want to figure out in the next year is where the line is between "agent that helpfully drafts" and "agent that quietly makes things worse." Most of my open-source work right now is variations on that question, narrating diffs, previewing changes, keeping the human in the loop without drowning them in approval prompts. If you're working on similar problems, or you want a senior engineer who'll happily ship Node and go yak-shave a ReAct loop on the same Tuesday, I'd like to hear from you.
