Intelligence that fits our country

Most AI used in Australia is built overseas, tuned for someone else's defaults and served from someone else's data centre. Australia should also have capable models of its own: open, efficient enough to run locally, and built with Australian English and Australian context from the start.

Data Craftsman is an Australian research and engineering lab building those models. We start small, release everything openly, and publish what we measure.

Early days, built in the open

Our first release is a set of Australian English voices, and our first language model is in training now. Every model we ship comes with open weights and a model card that sets out what it can and can't do.

Compute decides how big we can go. If you'd like to help us train larger Australian models, we'd love to hear from you.

Partner with the lab
Michael Jennings

M. JENNINGS · NEWCASTLE, NSW

Michael Jennings

I'm an engineer with fourteen years across engineering, data, infrastructure and AI in large Australian enterprise and government organisations. I've built enterprise scale AI platforms that run entirely offline, on premises, with no data leaving the network and thousands of users.

I started Data Craftsman to move from deploying other people's models to training our own.

Model training. Post training, distillation, quantisation and evaluation of open weight models, and speech models trained from public domain audio.

Local inference. GPU sizing, serving engines and the infrastructure needed to run models well on hardware you own.

Production experience. I've built platforms that run in production on premises, so I optimise for models people can deploy.