About
Ut ab ordine chaos, sic ab absurditate veritas.
As from order, chaos; so from absurdity, truth.
PhD studying evals under randomness with GNNs.
Particularly interested in evaluation frameworks, turning research into useful production ML systems, and the intersection of randomness and model performance.
Reach out: will dot leeney at gmail dot com
Timeline
AI Research Engineer
October 2025 - PresentFinetuning small to medium language models for tool calling. Generating synthetic training data using branching methods, working with Unsloth, vLLM, and Modal. Built semantic search across 10k+ provider actions and explored GPU memory optimisation principles for efficient training.
AI Engineer
July 2025 - October 2025Researched and built an AI agent that translates provider errors into clear, actionable resolution steps. Took an evals-first approach from day one — designed the evaluation pipeline, created custom tools, integrated into the product, and built monitoring dashboards.
ML Engineer
Built predictive models for early disease detection using remote patient monitoring data for AMD macular degeneration patients. Developed time-series training partitions for handling variable-length patient sequences. Proud to have built an AI anomaly detection system supporting the mission of preventing avoidable blindness.
PhD in Artificial Intelligence
2020 - 2024Researched best practices for evaluations and evaluating under randomness. Thesis: 'Unsupervised Graph Neural Networks' - focused on evaluation methodologies and model comparison. Defined metrics for quantifying model performance under uncertainty, with applications in graph neural networks and beyond.
MEng Engineering Mathematics (1st with Honours)
2016 - 2020Applied mathematics to real-world engineering challenges through an interdisciplinary lens. Core focus on mathematical modelling, data science, and machine learning fundamentals. Master's dissertation explored biologically-inspired RNN Hebbian learning rules for decision-making processes. The program's emphasis was on on bridging pure mathematics with practical engineering applications.