Warning: Undefined variable $publishedDate in /srv/users/serverpilot/apps/production/public/wp-content/themes/adapt/templates/single-post.php on line 18
In this interview, Dan Hart, General Manager AI at Sydney University, talked about creating CurricuLLM to personalise learning for every student.
Dan Hart is using large language models (LLMs) to support Australian public-school students with CurricuLLM, a new AI tutor that embeds the Australian school curriculum directly into its system.
Building on his earlier project, NSWEduChat, the tool delivers personalised learning by mapping student progress against more than 20,000 curriculum outcomes in real time.
Unlike generic models, CurricuLLM does not grade or assess students.
“It’s there so when they start a conversation, we can load up where they’re up to,” Dan says.
“Every conversation begins with the student’s individual context, so it’s truly personalised learning for every student.”
Helping teachers focus on teaching
The tool is also designed for teachers, requiring less prompting than models like ChatGPT. Educators can ask broad questions and receive curriculum-aligned responses instantly.
“It removes the need for teachers to learn how to prompt effectively or manage the context of the LLM conversation.”
While several public schools are interested in adopting CurricuLLM, licensing remains a challenge due to Australia and New Zealand’s 12 different curriculums.
Dan’s team is now finalising agreements with each state.
Lessons from EduChat
CurricuLLM builds on the success of NSWEduChat, which the NSW government rolled out to all public school students in 2025.
During its two-year development, Dan’s team tracked the steady improvement of AI performance, particularly in maths and science. By launch, the model performed accurately across almost every K–12 subject.
EduChat also proved valuable for students who lacked confidence to ask questions in class or who spoke English as an additional language.
Teachers noted stronger engagement and improved understanding across diverse classrooms.
Built for education, not automation
Dan believes educational AI must be purpose-built. “EduChat has Socratic questioning built in, so it’s next to impossible to cheat,” he says.
The platform is anchored to the curriculum and includes strict content filtering to maintain a safe learning environment.
Now as General Manager of Automation and Innovation at the Sydney University, Dan is leading projects focused on agentic process automation that streamline back-end operations using AI.
“Technology should support teachers and students, not replace them.”
Read our full interview with Dan Hart for more on CurricuLLM’s development, EduChat’s rollout, and the next chapter of AI in Australian education.