We are currently working on the following projects/questions:
1) Monash-MedCR: Permissive and Synthetic Medical Case Reports
Medical case reports (MCRs) represent a rich narrative clinical reasoning data source which may be invaluable in the training and evaluation of medical LLMs for self-dialogue and multi-agent communication. This project aims to release a large-scale HuggingFace dataset with over 250,000 permissive MCRs, as well as >1M synthetically generated low-similarity MCRs from high-quality seed MCRs.
2) CLARITY: RAG for Clinical Guidelines at Monash Health
MDs face an overwhelming proliferation of clinical practice guidelines. In collaboration with Monash Health CMIO Alex Duong, this project aims to deploy a RAG-LLM system with hybrid dense-sparse methods to integrate over 5,000 clinical guidelines and create an assistant capable of providing timely, concise and up-to-date guidance to over 15,000 clinicians.
3) ImmunoFM: A Gut Immunology Foundation Model
Foundation models (FMs) represent a powerful tool to model complex biological interactions. In collaboration with the Deputy Head of Immunology at Alfred Health, Prof. Benjamin Marsland, this project aims to build a multi-omics FM for downstream expression prediction. This project also provides an opportunity to understand whether FMs can capture biological dynamical systems, a common artifact of immunological circuits.
4) Beyond Attention
It has been seven years since attention has revolutionized the AI field, and we need to start thinking about the next paradigm shift. This exploratory project aims to draw inspiration from how children and adults learn to systematically examine potential architectures for future paradigm shifts in AI.
5) Medical AI Agents
AI agents and text-to-action represent powerful new paradigms for clinical practice. This project aims to identify the most effective agentic workflows for clinical medicine and understand how they may be integrated into clinical workflows, for example, through UI-based control of EHRs or clinical tool use.
6) Medical AI Governance and Infrastructure
Significant questions remain on how exactly to regulate and reimburse for medical large language models (LLMs). This project aims to provide regulatory, reimbursement and health economic recommendations for medical AI infrastructure, such as the FDA and Medicare. This project also aims to assess potential fairness alignment strategies for medical LLMs.
7) A Unifying Marker of Aging
The aging field requires a reliable marker of cellular aging. This exploratory project aims to assess the potential of spectral graph theory and graph transformers in providing a unifying marker of aging across biological data.