I am a critical care physician (medical intensivist) and researcher based in Montreal, Canada. My work focuses on the intersection of artificial intelligence, clinical decision support, and data security. I am passionate about leveraging technology to improve patient care and enhance clinical decision-making.
I earned my MD from McGill University in 2015, where I completed residency training in Internal Medicine. I subsequently obtained subspecialty certifications in General Internal Medicine and Critical Care at the University of Montreal. Following my residency, I pursued additional training in Biomedical Engineering at Johns Hopkins University, as well as a postdoctoral research fellowship at the Reasoning and Learning Lab at Mila. I now practice as an intensivist at CHUM (Centre Hospitalier de l'Université de Montréal), where I also teach and supervise trainees. In 2018, I co-founded Pathway Medical, an AI-driven platform to help clinicians practice evidence-based medicine, and currently serve as its Chief Medical Officer.
Pathway is an AI-enabled clinical decision support platform that uses knowledge graphs and large language models (LLMs) to provide evidence-based answers to clinical questions in real time. It leverages Retrieval-Augmented Generation (RAG), Chain-of-Thought (CoT) reasoning, and structured medical knowledge graphs to produce answers with transparent reasoning that enable healthcare professionals to better understand, verify, and trust AI-generated guidance. Pathway's models show state-of-the-art accuracy on standardized medical assessments like the USMLE and MedQA.
I have been involved in research around federated learning and artificial intelligence in healthcare. I co-led the development of the CODA platform – an open-source infrastructure for federated data analysis and machine learning on distributed healthcare data. CODA enables secure, collaborative research across institutions without centralizing sensitive patient data, and we detailed this work in the Journal of the American Medical Informatics Association in 2024. CODA integrates modern healthcare data standards (FHIR and DICOM) and provides built-in, no-code data visualization tools, explicitly controlled data disclosure mechanisms, and scalable FL capabilities. In a proof-of-concept study, we successfully deployed CODA across eight public hospitals in Quebec, Canada, and enrolled over 1 million patients, showcasing its ability to scale and manage extensive multi-modal healthcare datasets. Technical feasibility studies revealed that major challenges included data conversion from legacy systems and securing dedicated IT resources. The platform was validated through end-to-end multi-modal FL testing using public datasets (MIMIC-IV and MIMIC-CXR), demonstrating comparable performance to centralized pooled analysis in mortality prediction tasks. CODA's design emphasizes robust privacy measures, secure data handling, and accessibility to non-technical users, fostering broader adoption of federated learning methodologies in healthcare research. Ongoing efforts aim to enhance platform functionalities, further validate predictive models in real-world scenarios, and expand tools supporting data migration from legacy systems to modern standards.
I have a strong interest in clinical informatics and medical imaging, particularly in building tools that bridge data to bedside decision-making. With Christophe Marois, I co-developed the initial version of CoreSlicer, a web-based toolkit for analytic morphomics, which enables clinicians to extract body composition biomarkers (like muscle and adipose tissue measures) from CT scans. This software, described in BMC Medical Imaging, allows even non-experts to perform image analyses for frailty and risk assessment.
Improving the care of frail and elderly patients has been a recurrent theme in my research. In the BICS study, we showed that bioimpedance phase angle is a robust biomarker of frailty and a predictor of mortality in older adults undergoing cardiac surgery. Published in the Journal the American Heart Association in 2018, this work showed that patients with lower phase angle (indicative of poor muscle quality) had significantly higher post-operative morbidity and mortality. I have also studied sarcopenia (loss of muscle mass) in cardiovascular patients. I co-authored a report in JACC showing that low muscle area on imaging correlates with worse outcomes after transcatheter aortic valve replacement.
During my undergraduate studies, I was involved in the creation of Syme, a social networking platform established around 2013, developed in direct response to rising concerns about privacy infringements and mass surveillance highlighted by Edward Snowden's disclosures about NSA activities. Committed to providing a genuinely private alternative to mainstream platforms like Facebook and Twitter, Syme employed end-to-end encryption, ensuring that all content—including posts, messages, and images—remained accessible only to intended recipients. As a zero-knowledge platform, Syme's servers were incapable of accessing or viewing user-generated content, reinforcing user privacy and data security. The platform also featured advanced privacy controls similar to Google+ Circles and Facebook Friend Lists but reinforced through robust encryption mechanisms. Initially targeted at academic and activist communities, Syme provided a secure environment for sensitive communication and collaboration.
While ultimately shut down following the incorporation of encryption features in mainstream social network applications, Syme was a pioneering figure in the evolution of privacy-focused social media platforms.
I am deeply involved in academic medicine and teaching. I serve as Adjunct Professor of Medicine at the University of Montreal's Faculty of Medicine, where I give a class on regulation of body fluids. I supervise medical students and residents in the ICU and teach principles of evidence-based critical care. Earlier in my career, I was a Teaching Fellow at Harvard University (2013), which gave me formal experience in pedagogy. I have mentored trainees on research projects, including guiding medical residents and graduate students in data science initiatives within our ICU research group. My dual role as a practicing intensivist and researcher allows me to educate colleagues on applying AI and informatics in clinical settings – for instance, I often give lectures on machine learning in healthcare and have organized workshops on using the CODA platform for collaborative research. I also contribute to continuing medical education, delivering sessions on up-to-date critical care guidelines and the use of decision-support tools like Pathway. Teaching and mentorship are rewarding aspects of my career, and I'm committed to training the next generation of clinician-scientists.
(For a complete list of publications and software projects, please refer to my Google Scholar and GitHub profiles, respectively.)
I am always happy to connect and collaborate. Feel free to reach out on Twitter/X (@LouisMullie) for any inquiries related to my work or potential collaborations.