Australian domain data
Australian legislation, regulations, case law, domain-specific corpora, English spelling, idioms, and civic context.
SCX MAGPiE
Thinking like an Australian.
Project MAGPiE is a sovereign approach to reasoning alignment in large language models. Instead of forcing models to sound Australian through prompts, the project embeds Australian reasoning patterns, legal frameworks, and cultural context into the model's thinking process.
The challenge
Generic LLMs often reason from foreign assumptions.
Traditional approach
Problem
MAGPiE direction
US-centric training data
Foreign legal and cultural defaults
Australian context from the ground up
Massive system prompts
Brittle behavior and token overhead
No extensive prompting to act Australian
Surface localization
Sounds local but reasons elsewhere
Native understanding of ASIC, APRA, law, language, and culture
How it thinks
Reasoning aligned with Australian context.
Reasoning process
The model evaluates options through Australian legal, cultural, business, and institutional frameworks.
Natural output
Outputs are designed to align with Australian expectations without relying on long prompts to fake locality.
Financial services
ASIC guidelines, tax law, superannuation, and compliance workflows without extensive prompting.
Legal and compliance
Australian case law, legislation, precedents, and legal-domain context.
Government services
Australian government processes, policies, citizen services, healthcare, education, and local operations.
Research link
MAGPiE is evaluated through Australia 1k.
Australia 1k assesses cultural norms, language, history, media, and institutional alignment.