Work

Machine learning + physics researcher · UCL · QTNM collaboration

Working within the QTNM collaboration (twenty researchers across five UK institutions) on precision neutrino mass measurement via cyclotron radiation emission spectroscopy. Designing and benchmarking deep learning architectures — CNNs, Transformers, diffusion models — for signal detection and reconstruction in noisy time-series data. Applying simulation-based inference for posterior estimation. Integrating models into the live analysis pipeline.

Remote, UK

Risk-aware transaction profit model using gradient-boosted trees, trained on two years of booking-level data with a claim-weighted loss. Causal lift estimation pipeline using a doubly-robust estimator (Chernozhukov et al. 2018) to quantify the profit impact of counterfactual price changes before any change went to pilot.

• Segmented repricing mechanism: raise prices on the high-risk half, leave the rest unchanged. Delivered as a cache-key pricing table for the live pilot.

Cambridge, UK

Transformer-based automatic speech recognition for 100M+ edge devices, within tight latency and memory constraints. Achieved a 10% relative word error rate reduction. Adopted as the company’s forward ASR strategy, reducing reliance on third-party vendors and laying groundwork for on-device conversational AI.

Manchester, UK

Reinforcement learning agents (A2C, PPO) for inventory management, with custom sustainability penalties written into the reward function. The work contributed to a sponsored PhD studentship at the University of Manchester.

Supervised student machine learning projects across physics and engineering applications. Projects have included CNNs for land-use segmentation, LSTM weather forecasting, NLP political rhetoric analysis, and 3D CNN neutrino event reconstruction.

Conservation editorial and analysis site. Data-driven writing on ecology, wildlife, and the numbers behind the natural world.

Personal site and writing home. Built with Astro on GitHub Pages, obsessively trimmed: no framework overhead, no build complexity, just content and craft.

University College London

QTNM collaboration. Thesis submission target: September 2026.

University College London

80%. Modules included High Performance Computing (CUDA), Practical Machine Learning, Computational and Quantitative Finance, Advanced Quantum Theory, Statistical Physics.