Australian AI Research & Engineering

Capable, efficient AI that Australia owns.

Data Craftsman is an Australian AI research and engineering lab. We train open models for Australian English, Australian context and hardware you own.

See our research Partner with us

01 / Mission

Our mission is to make AI work for Australians

Our long term aim is an Australian open weight model that small businesses and organisations can run on hardware they already own. We are starting with smaller models and publishing what we learn along the way.

Open weights. We release model weights for anyone to download, inspect, run and build on. We publish our code on GitHub.

Runs on hardware you own. We design for efficiency from the start so a small organisation can run our AI in its own building without sending a single token outside.

Evidence before claims. We publish results with the method used to get them, including where a model falls short. We don't say something works until we have measured it.

Built for Australia. We train Australian English, Australian context and Australian defaults into the model itself.

02 / Latest release

Australian voices, free to use and fully offline

piper-en_AU-librivox-medium

Ten Australian English voices for Piper, the open source text to speech engine. Trained on public domain LibriVox recordings, small enough to run on a CPU, and fully offline.

~70M parameters
77 MB
Runs on CPU
CC BY 4.0

View on Hugging Face

Text to speech

03 / Now training

What we're training now, and what comes next

Training has begun on Australis Black Opal, the first model in our Australis text family. Next come open Australian models for decision making and voice and, with enough compute, Australis Olympic: a flagship pretrained from scratch.

01 / 04

01 / Training now

Australis Black Opal

The first model in our Australis text family: a recurrent language model, built as a 26B mixture of experts model for long documents, code, tools and Australian workplace tasks.

Gemma 4
Recurrent depth
Australian English
Text + image
Open weights
Apache 2.0

Base model: google/gemma-4-26B-A4B-it
Architecture: Gemma mixture of experts. Recurrent depth must beat the native model at matched compute before it ships.
Precision: NVFP4 deployment build. We will publish the final mixed precision map after measurement
Built to handle: Long documents, code, tools and real Australian workplace tasks
Deployment: One 32 GB+ Blackwell class workstation GPU
Quality target: An independent evaluator must measure ≥20 on the AA Intelligence Index. Australian benchmarks cover lexis, spelling, culture and knowledge, and the model must retain general and multimodal capability.
At release: Open weights, per task results including failures, and a technical report covering data, training and trade offs

02 / Up next

Umpire

Our open Australian decision model. Inspired by TypeSafe AI's Jev, it is designed to make the call: choose from a fixed set of options, fast, with Australian context, institutions and logic.

Open source
Australian logic
Decision reasoning

Purpose: Make structured decisions using Australian assumptions and context rather than relying on overseas defaults
Approach: Build an openly available Australian counterpart to models such as Jev

03 / Planned

Lyrebird

Our next generation Australian voice model: higher quality, to succeed our first Piper release and move beyond its 29.6 hour training run.

Text to speech
Australian English
Open source
Piper successor

Starting point: Lessons from piper-en_AU-librivox-medium, our first voice model, trained for 29.6 hours
Goal: A next generation voice system with more voices, a larger training effort, better voice quality and potentially different base

04 / Seeking compute

Australis Olympic

The flagship of the Australis family, named for the Olympic Australis, the largest gem opal ever found. A new language model pretrained from the ground up under a new architecture. This is the long term project, and it depends on funding and access to substantial GPU compute.

Pretraining
New architecture
From scratch
Funding required
GPU compute required

Research direction: Design and pretrain a language model rather than adapting an existing base model
Dependency: External funding, GPU hardware or cloud compute sufficient for pretraining
Status: Work on the architecture and training programme will begin when we secure the required resources

All research and releases

04 / Partner with us

Help us pretrain an Australian model from scratch

The lab is self funded, and compute limits the size of what we can train. Our long term goal is a language model with a new architecture, pretrained from the ground up. If you can offer funding, GPU time, hardware or cloud credits, we'd like to hear from you.

Partner with us