Clemens R1, Unsworth K2
1Australian Research Data Commons (ARDC), University of Queensland, St Lucia, Australia, 2Australian Research Data Commons (ARDC), Monash University, Caulfield East, Australia
Biography:
An ecologist with experience in ornithology and spatial modelling, Rob identifies skills gaps, formulates skills strategies and co-designs skills-related resources, so that the data and analytics delivered through the ARDC Planet Research Data Commons can be used to their full potential by earth and environmental science researchers.
Abstract:
Traditional learning systems often lack the agility to address highly specialised research training needs, or unique individual questions. This presentation demonstrates how trainers with limited software engineering expertise can rapidly develop and deploy tailored, self-hosted Artificial Intelligence (AI) tutors. We present the end-to-end workflow behind the "Australian Dataspace Educator," a proof-of-concept application built utilising open-source tools and hosted on Australian Research Data Commons (ARDC) infrastructure.
The core problem addressed is ensuring generative AI remains contextually accurate and pedagogically useful rather than simply transactional. We outline a Retrieval-Augmented Generation (RAG) architecture grounded in curated materials, employing the ARDC skills framework to tailor user journeys based on specific learner personas and required competencies. When the localised knowledge base is insufficient to answer a query, the system falls back to a primary generative Application Programming Interface (API) to ensure continuous support.
Attendees will view a live demonstration of the application alongside an analytical dashboard derived from user logs. This dashboard maps gaps in the training curriculum, highlights common queries, and identifies AI imperfections. We also share a machine-readable workflow document hosted on the Digital Research Skills Australasia (DReSA) registry, encouraging other trainers to reproduce this architecture and register their own educator applications.
Ultimately, this workflow empowers trainers to scale accurate responses to highly specific user questions. We conclude by exploring future enhancements, such as structured knowledge graphs and lightweight learning management systems, demonstrating that bespoke and safe AI training ecosystems are now fully achievable for the research community.
