Zelo
Making AI usable without the enterprise tax — local agents, custom automation, practical deployments.
Occam's razor
the simplest explanation tends to win
01 / PERSON
“I was just a child, walking in the wild, learning my name”
— Skegss
Data scientist & software developer in Melbourne. I build systems that hold weight — AI products, analysis pipelines, and tools meant to ship.
Gradient descent
“Going down the only road I've ever known”
— Lynyrd Skynyrd
Take the slope. Step. Repeat. That's how anything gets built.
Making AI usable without the enterprise tax — local agents, custom automation, practical deployments.
Digging into human vs chimp brain RNA-Seq — patterns that make biology legible through ML.
Signal vs. noise
“I can see clearly now, the rain is gone”
— Johnny Nash
Most of the data is texture. The red shape is the point.
Implicit ALS on AIS tracks: vessels × ocean cells, fishing hours as confidence — not ratings. Same factors recommend MPA-legal grounds, flag IUU-like outliers, and forecast seasonal effort. ~20× a popularity baseline.
Not cloudy-or-sunny — a stability score for where Melbourne light will hold still. Contour bands over the city, live 1/3/6-hour windows, ranked shoot spots.
Hourly CBD pedestrian forecasts by precinct — sensors joined to weather, calendar, and land use. Measures how each street recovered after lockdowns, not a citywide average.
Webcam swipe launcher. Optical flow watches a gesture band above the keyboard; while armed, a horizontal swipe opens an app. The whole UI is an ON/OFF button.
Overfitting vs. generalization
“Stuck in a moment and you can't get out of it”
— U2
Memorize the training set and the next point cracks you open.
The model that generalized was never the prettiest fit on the training set.
If it only works on the notebook, it doesn't work. New points are the point.
Constraints — time, clarity, honesty — keep the system from memorizing noise.
Bias–variance tradeoff
“I've looked at life from both sides now”
— Joni Mitchell
Too sharp misses the world. Too soft misses the edge. Hold both.
Affordable local AI agents. Custom automation, scraping, and intelligent workflows for people who refuse enterprise price tags.
Retail + technical consultation on Meta platforms — turning product questions into clear next steps.
Shipped ML solutions for real datasets — processing pipelines, predictive models, and evaluation loops.
NEXT / CONVERGENCE