Research / Selected studies
Medical AI at MIPSLAB, ongoing honours work at ETHEREAL Lab, and earlier ML studies. The questions, my contributions, and the evidence behind each project.
MIPSLAB, University of Calgary
Can tumour segmentation support downstream cancer staging?
Worked with PET/CT scans to segment head and neck tumours, extract imaging-derived features, and test whether those features could support downstream cancer staging.
Contributed to PET/CT preprocessing and the tumour-segmentation workflow.
≈0.7 · Project segmentation Dice
ETHEREAL Lab, University of Calgary
How does a communication-partner prompt relate to participant state?
My honours research looks at how communication-partner prompts relate to what a participant is doing, with the longer-term goal of understanding when different kinds of prompts are useful.
Clean and structure coded interaction data.
Ongoing analysis of prompting strategies
Can a webcam turn a hand gesture into useful interface feedback?
Who is more likely to subscribe after a bank marketing campaign?
Can a dense neural network recognise a handwritten digit?
Open an area for a little more context.
At MIPSLAB I worked on PET/CT segmentation and feature-based staging. I’m interested in how preprocessing and evaluation affect what those models can tell us.
HECKTOR / segmentation & stagingMy honours work at ETHEREAL Lab studies prompt content alongside participant interaction state. The analysis and modelling pipeline is ongoing.
AutomaViC / ongoing honours researchSilentSpeak connects MediaPipe hand tracking, a gesture classifier, and a browser interface. I’m interested in the engineering between a trained model and usable real-time feedback.
Recognition & interfacesFeature engineering, class imbalance, and model evaluation recur across my medical-imaging and earlier tabular ML work. I want the metric to answer the question the project actually asks.
Prediction & evaluationSwiftApply runs Gemini Nano on-device, while SilentSpeak runs gesture inference with TensorFlow.js. Browser APIs, local data handling, and model availability shape what the software can do.
Local inference & application integrationI’m interested in tools where a person’s context and judgement remain part of the system: prompting research, local browser AI, and reviewable financial data workflows.
Interaction, context & review