About

I’m Jason Matthew, an engineer in Brisbane, Australia. I build production AI systems and write about the engineering underneath the hype.

For the last couple of years most of that has been applied-AI infrastructure: retrieval pipelines running in production, an S3 Vectors integration built through the AWS preview (named as a launch partner on the product page), and ingestion across more than a million pages. The interesting part is rarely the model. It’s getting a non-deterministic system to hold up against real data, real users, and the edge cases nobody wrote down.

Building Is How I Think#

By night I ship side projects. Skopia, privacy-first analytics you self-host on your own Cloudflare account. A LoRA fine-tune that beat every baseline on form-abandonment classification. A Cloudflare-native RAG that answers tabletop rules and cites every ruling so you can check it. A pipeline debugger for the terminal, written in Rust. Not because I have to, but because building things is how I think, and because staying hands-on is the only way I trust my own architecture calls.

I’ve also led engineering teams across AI, Search, and platform work. That taught me how organisations actually ship and where they stall, and it made me a better engineer. But the code is where I do my best work: setting the architecture, building the tools, and shipping the systems that move the number.

What I’m Working On#

Right now I’m going deeper on the parts of applied AI that are still hard. Evals you can point at your own RAG, and continual-learning for retrieval systems that drift over time. It’s in progress, not shipped, and I write it up as I go.

Outside Work#

When I’m not at a keyboard, I’m usually on two wheels or two feet. I race road and ride mountain bikes, with a lifelong goal to clear A-line at Whistler Bike Park, a dream I’ve had since visiting Canada as a kid. I also run marathons, and I won’t stop until I beat Will Ferrell’s 3

. He’s the benchmark. Don’t ask why.