Projects
Things I built because I wanted them to exist. Two of them ended up with real users, which is still the strangest and best part.
01 · OPEN SOURCE · JAVA
Ngrok‑LAN
A Fabric mod that lets you share a singleplayer Minecraft world with friends anywhere, by clicking one new button in the pause menu. It opens the world to LAN, finds a free local port, and tunnels it out through Ngrok, so nobody has to touch a router or forward a port. The generated address lands in chat as a click‑to‑copy link.
The interesting parts were the ones players never see: starting the tunnel off the main game loop so the client does not hitch, mixins into the LAN screen and player manager, a config for auth token and datacentre region, and enough error handling to explain Ngrok's session limits in plain language.
DOWNLOADS34
GITHUB STARSFABRIC · MIXIN
QUILT PORT IN PROGRESS
02 · WEB APP · AI
Chewmetric
Calorie and macro tracking for people who would rather eat than count. Describe a meal in words, photograph your plate, scan a barcode, or photograph a restaurant menu, and it decomposes the whole thing per ingredient: calories, protein, carbs, fat, fibre. The menu scanner ranks dishes by how well they fit whatever macros you have left for the day.
Behind the conversational surface it is a lot of deterministic bookkeeping: barcode lookups against Open Food Facts, percentage maths that always sums to 100, a coach that keeps your allergies and preferences across sessions, and recipe and meal‑prep generation to fill the gaps. Installable as a PWA, free, offline first.
MULTI-MODAL INPUTPWA
OFFLINE FIRST
03 · iOS · MACHINE LEARNING
Morbas
A practice tool for musicians, published on the App Store. Morbas pulls a song apart into vocal, drum, bass and instrument stems, then lets you slow it down or shift its pitch, so you can play along with only the parts you need. I built it because learning a drum line by ear is much easier when you can mute everything else.
Separation models are too heavy for a phone, so inference runs on AWS: custom ECR images on ECS, scaling elastically with demand, with the surrounding workflow handled by Lambda. The app stays thin and the cost stays close to zero when nobody is practising.
VOCAL · DRUM · BASS · INSTRUMENTAWS ECS
ELASTIC INFERENCE