Some of these are research code. Some are software that people use on a Tuesday morning to do their job. The second kind teaches me more.
Author and maintainer · 2026
The reference implementation of the HARMONY paper: the local Cox proportional hazards trainer, the peer reputation exchange, the adaptive lambda controller, and the full SEER preprocessing pipeline.
Outcome. MIT licensed and public since May 27, 2026. Reproduces the JBHI results end to end, so they can be rechecked rather than taken on faith.
Built with Python, PyTorch, lifelines
Entrepreneurial Lead · 2022 to present
A federated survival analysis platform for personalized oncology treatment recommendations, and the NSF I-Corps venture built around it. The interviews changed the product more than any experiment did. The team completed the Great Lakes Hub W23 Medical Jumpstart course and the NSF National I-Corps program. I drafted the NSF TTP-P proposal submitted in January 2026, which was returned without review, and led the rebuttal.
Outcome. NSF National I-Corps completed with a $50,000 award, after 157 customer-discovery interviews with clinicians, researchers, and industry people.
Built with Python, Federated learning, Survival analysis
Designer and sole developer · v1.3.0 shipped September 5, 2026
A production oncology EMR and clinical decision support desktop application, built for a practicing oncologist and used in their clinic. It does OCR on scanned documents, offline dictation that never sends audio anywhere, TNM staging, ECOG and Karnofsky scoring, and BSA calculation. It ships as a signed macOS installer. In August 2026 it became the product foundation for SmarTreatMate.
Outcome. In clinical use, with 466 automated tests across seven suites: a byte-for-byte data-preservation suite for upgrades, and a date engine that round-tripped 117,973 dates with zero failures. That is the kind of test you write when the file is somebody’s medical history.
Built with SQLite, OCR, Offline speech recognition, macOS packaging, Automated test suites
Sole developer · 2026, 165 hours
A cardiovascular risk screening platform for community health workers, built for a cross-university field study. It is offline first, because the people using it are often in places with no connection. Ethics approval shaped the data-protection constraints, and the test harness includes negative controls.
Outcome. Delivered as a mobile progressive web app and a native Android client sharing one backend.
Built with Python, SQLite, Android (Kotlin), Progressive web app
Developer · 2023 to present
A portable AI clinical assistant testbed running on an NVIDIA Jetson AGX Orin. It virtualizes healthcare centers so clinical scenarios can be tested in real time.
Outcome. Where the survival and federated models stop being a cluster job and have to run on hardware small enough to sit in a clinic.
Built with NVIDIA Jetson AGX Orin, Python, PyTorch
Author · 2022 to 2026
CoSurv covers federated Cox proportional hazards with clustering, trust and noise simulation, and missingness. FL-via-GT does game-theoretic peer selection. DyFeC and an IFCA implementation cover clustered federated learning. StateESR is the code for the CHASE 2023 paper.
Outcome. The code behind the papers, kept runnable so the results can be rechecked.
Built with Python, PyTorch, SLURM
Manager, organizer, and director · 2018 to 2020
A project with the SIGNET lab to review, standardize, and streamline how a hospital actually runs: patient flow, medical care, physical resources, and the internal systems holding it together.
Outcome. An RFID real-time location system for the hospital, which I managed and developed.
Built with RFID (UHF), C#, .NET, SQL Server