Australian AI BioHub
Preview site
This site is in development and not yet public. Enter the passphrase to continue.
Not quite — try again.
For access, contact biohub@adelaide.edu.au
Australian AI BioHub
Preview site
This site is in development and not yet public. Enter the passphrase to continue.
Not quite — try again.
For access, contact biohub@adelaide.edu.au
Australian Institute for Machine Learning · Adelaide University
An ecosystem for AI-enabled biological research, engineering, innovation and industry collaboration.
Foundational AI research, applied inside a health system — with methods depth, people who work across both sides, and access to live SA Health projects.
01
Australian and global bioresearch companies face critical barriers to AI adoption: talent scarcity and data access limitations. The demand for AI-skilled bioresearchers far exceeds supply, with companies struggling to recruit and retain professionals who understand both biological systems and machine learning applications.
Bioresearchers need AI expertise to stay current with rapidly advancing research methodologies, while companies require a single point of access to AI expertise and data to accelerate R&D without building capabilities in-house. Without systematic training and collaborative data platforms, Australian companies will struggle to compete with international counterparts who have better access to AI talent and integrated datasets.
01 / METHODS DEPTH
Five method themes with a clinical test: multimodal AI across images, text, sensors and omics; explainability clinicians can interrogate; generalisability across sites and scanners; small and noisy data; and causality. Published at NeurIPS, ICML, CVPR and ICCV — not applied AI assembled from other people's models.
02 / PEOPLE
Industry PhDs, embedded work on live projects and professional training build researchers fluent in both biology and machine learning.
03 / ACCESS
Live deployments inside SA Health, with the ethics, consent and Aboriginal-led data governance already running. Patient data cannot leave the network, so the work happens inside.
02
Three systems trialled or live in SA Health, with the AI built here. Figures last checked September 2026.
2020–2025
12 hospitals
The first AI-based randomised clinical trial for chest pain: decision support embedded in emergency-department workflow across twelve South Australian hospitals, six metropolitan and six rural, with 14,131 patients enrolled. The AI did not reduce cardiovascular death, MI or unplanned readmission at six months. It was safe, and it changed practice: patients not classified as type 1 MI were 47% less likely to undergo invasive angiography, and those classified as type 1 MI were more likely to receive guideline-directed therapy. Late-breaking research, ESC Congress 2024.
AI built by the group · NHMRC $1.23M, led by Flinders · SA Health · Siemens Healthineers · InterSystems · CEIH
2024–
3 emergency departments
A sovereign ambient AI clinical scribe, founded and led here. The entire pipeline runs inside SA Health's private cloud, because SA Health policy prohibits patient data leaving the network — which made commercial scribes unusable and building the only path. In proof-of-value deployment across three emergency departments with more than 50 clinicians using it, alongside a dedicated program measuring and correcting speech-recognition bias for Aboriginal patients under Aboriginal-led governance.
SA Digital Investment Fund $1M · MRFF extension $1.9M · CALHN · CEIH · Digital Health SA
2023–2028
SA Pharmacy
Automated, nationally scalable medication-safety review over My Health Record data. The largest deployable AI project in SA Health, with the AI components built by the group.
MRFF NCRI $2.92M · SA Pharmacy
2023–
CALHN
A real-time dashboard for cardiovascular MBS and PBS insights, letting health services see unwarranted variation at population scale.
Supported by the Hospital Research Foundation
03
Mission
To harness AI and South Australia's deep methods expertise at AIML to accelerate discoveries in biological research, enhance personalised solutions across industries, and create scalable innovations for global challenges.
Vision
To establish Australia as the epicentre for AI-driven innovation in the biological sciences and engineering, translating discoveries into tangible solutions for healthcare, agriculture, environmental sustainability and industry.
04
Across research, education and commercialisation.
Help companies and academia develop and deploy AI tools for faster, more accurate analysis of complex biological and engineering data.
Upskill researchers and partners through industry PhDs, embedded projects and professional training, so they can use AI tools effectively to advance their work.
Collaborate with industry and university partners to transform scientific innovations into market-ready products.
Build networks of scientists, engineers and medical professionals around major scientific challenges.
05
Our interdisciplinary approach targets the most pressing challenges in biosciences and engineering. One area is operating, four are active, four are staged.
Clinical AI built inside the health system rather than bought from outside, evaluated with randomised trials measuring system outcomes, and governed with consent, ethics and Aboriginal-led data arrangements already in place. Three systems so far: RAPIDx AI, AUScribe and AutoMedic.
A/Prof Johan Verjans · Dr Zhibin Liao · CALHN · SALHN · SA Pharmacy · CEIH
Deploy machine learning to decode complex omics data, accelerating the discovery of genetic markers, disease mechanisms and therapeutic targets.
[TODO: lead]
Leverage computer vision and foundation models to automate analysis of histopathological images, improving diagnostic accuracy, throughput and consistency — and industrial process inspection.
[TODO: lead] · SA Pathology
Personalise treatment regimens using AI models that predict patient responses to therapies, improving outcomes and reducing healthcare costs.
[TODO: lead]
Support development of next-generation sensors and AI-enhanced devices for healthcare, environmental monitoring and industrial processes.
[TODO: lead]
AI-driven analysis of microbiome data to identify microbial consortia with therapeutic potential and expedite development of novel microbiome-based therapeutics. Pharma partners are already in place.
Needs a lead and a first grant round
Develop AI solutions for optimising agricultural productivity and sustainability through real-time monitoring of environmental factors and crop health.
Needs an anchor partner
Self-driving labs where AI proposes experiments, robotic platforms execute them and results retrain the models — plus field and clinical robotics for agriculture, biomanufacturing and care delivery.
Needs an anchor partner and a demonstrator
Quantum and photonic sensors produce faint, noisy, high-dimensional signals that only become useful with machine learning. We pair sensing capability with AI for signal extraction, calibration, drift correction and interpretation.
Needs an anchor partner; DST Group is the closest existing route
06
Our structured approach connects core AI domains with application areas for targeted innovation. Read down a column to see what a domain can draw on; read across a row to see the transfer argument — a capability, held by a named person, proven in one domain and moving to the next.
07
Real problems nominated by clinicians. Researchers and companies pick from this list.
Handover notes are inconsistent between shifts, and the gaps show up as repeated investigations. Structured capture at handover, in SNOMED CT.
Adherence differs between metropolitan and rural sites. Quantify the variation, then identify where decision support would and would not help.
Manual and time-critical. Where does automation reduce error, and where does it introduce new failure modes?
Measure and correct recognition bias, with Aboriginal-led governance and data sovereignty as requirements rather than considerations.
Slide review capacity is the constraint. Which cases can be safely ordered by a model, and what is the evidence bar for saying so?
08
A curated list with judgement applied — not a course. Every entry has an owner and a review date.
For clinicians
What these systems can and cannot do, without mathematics. The most common starting point.
Clinically framed: data, modelling and evaluation in care settings.
The reporting standard for prediction model studies. Read it before a vendor's evidence pack — it tells you what should be there.
Reporting guidance for early clinical evaluation of decision support — the stage most tools are actually at.
The reference document when a service is asked to adopt something quickly.
For researchers
Code-first. The fastest route from Python to working models for a domain researcher.
Interactive textbook with runnable notebooks. For filling gaps rather than reading end to end.
Protocol and reporting standards for clinical trials of AI interventions. Start here before designing one.
Working with real de-identified critical care data, including access and credentialing.
The standard single-cell workflow. Relevant to the multiomics work.
For patients and the public
Where these tools are used, what they do, and what a clinician still decides.
Where it goes, where it doesn't, and why our systems run inside the SA Health network.
A short list you can take to an appointment.
What it means, and how it shapes the way we build.
These pages do not replace talking to your clinician.
09
Open source only, grouped by task. The notes are the point — a list of links to well-known repositories helps nobody.
De-identification
Configurable PII detection and anonymisation. The default recognisers are US-centric — Medicare numbers, IHIs, URNs and Australian address formats need custom recognisers, and that work is where the accuracy actually comes from.
Clinical and biomedical NLP pipelines, useful as the entity layer under de-identification. scispaCy is trained on biomedical literature, not clinical notes — expect a drop on ward documentation.
Speech recognition
Whisper hallucinates confidently on silence and cross-talk, which is exactly what a consultation recording contains. Voice activity detection before transcription matters more than model size. Accent performance must be measured on your own population, not assumed from published benchmarks.
Running models inside a restricted network
Throughput-oriented serving for self-hosted models. In a private cloud the constraint is GPU memory and concurrency planning, not raw model quality.
Local prototyping, and useful for showing a governance committee that nothing leaves the machine.
Standards and terminology
Reference FHIR implementation. Structured output is what downstream tools stand on — a scribe that emits free text only is a dead end.
SNOMED CT terminology server. Licensing for SNOMED itself is separate from the software and needs checking early, not late.
Worth adopting if you will ever want to run the same analysis across more than one health service.
Imaging
Medical imaging deep learning framework with sensible defaults for 3D data and transforms.
Self-configuring segmentation. Beat it before you build something custom; most of the time you won't.
Evaluation, fairness and reproducibility
Subgroup performance assessment. The subgroups that matter are the ones the training data under-represents — measure before deployment, not after a complaint.
Experiment and data versioning. A governance committee will ask which model version produced a given output, and you need to be able to answer.
Licences and versions are confirmed at adoption. “In production” means we run it; everything else has been evaluated only.
10
Bringing together expertise in AI and the biosciences.
DIRECTOR
Clinical trials, digital health, medical AI
CO-DIRECTOR
Explainable AI, multi-modal AI
OPERATIONS & STRATEGY
Strategic planning and operational excellence
AIML faculty
Embodied AI
Robotics & intelligent automation
Causal AI
Computer vision
Generative AI
Open to clinicians and researchers
The Australian Institute for Machine Learning is home to one of the most respected applied AI engineering teams in the country. This award-winning team has been instrumental in translating machine learning research into real-world applications across healthcare, agriculture, advanced manufacturing and defence.
With deep expertise in software engineering, computer vision, natural language processing and scalable AI systems, the team bridges the gap between research and implementation.
Key capabilities
11
We partner with leading institutions and companies to drive innovation at the intersection of AI and the biosciences.
International industry
Local collaborators
Health institutions, defence and academia
12
The Hub contributes to Australia's leadership in AI-driven research and innovation across biomedical, environmental and engineering domains — improving outcomes, reducing the burden of disease, enhancing environmental sustainability, and driving economic growth through the development of new technologies.
The Hub is not just a research centre. It is a movement toward a smarter, healthier and more sustainable future powered by artificial intelligence.
Future goals
With international research centres, to expand our global reach and impact.
From government, industry and philanthropic sources, to scale our initiatives.
And commercialisation of AI innovations, to maximise real-world impact.
Through education, to build Australia's AI talent pipeline.
13
Two ways in: refine an existing collaboration, or bring a new partner-defined challenge.
Australian Institute for Machine Learning
Lot Fourteen, North Terrace, Adelaide SA 5000
biohub@adelaide.edu.au
+61 8 8313 3051
Overseen by the Hub Governance Committee, accountable to funders and Adelaide University.