Martin Wright

I am a researcher and engineer with a passion for solving complex problems at the intersection of machine learning, stochastics, and operations research. My work focuses on developing practical solutions that bridge theoretical advances with real-world applications.

Currently, I am exploring new methods in computer vision and deep learning, with particular interest in simulation, optimisation and meteorological applications.

When I'm not working as an engineer, I enjoy spending my time between surfing, tramping and raising my daughter.

domains

projects

Coastal Camera Image Processing

July 2026 · Computer Vision · Coastal Engineering · Image Processing

Python tools for Auckland Council's nine-camera coastal monitoring network. Transforms oblique time-lapse imagery into geo-referenced orthographic plan-view products using the pinhole camera model, robust homography estimation for frame stabilisation, and backward-warped rectification onto surveyed ground control points.

VAE for Hydrological Stochastic Simulation

March 2026 · Deep Learning · Stochastic Modelling · Hydrology

Generative model combining a variational autoencoder with adversarial training to produce realistic, physically consistent synthetic rainfall ensembles. Targeted at hydrological simulation and flood risk assessment workflows requiring large ensembles of plausible precipitation scenarios.

Stochastic Co-variant Rainfall Extension

Oct 2025 · Stochastic Modelling · Hydrology · Statistics

Methodology for extending short observed rainfall records using stochastic co-variate relationships with long-record atmospheric datasets. Preserves dependence structure and extreme-value behaviour, enabling return-period estimation beyond the observational record length.

Intelligent Semantic Caching

Feb 2025 · Software Engineering · Machine Learning · NLP

Semantic similarity-based caching layer for LLM API calls. Embeds queries and retrieves cached responses for semantically equivalent inputs, reducing redundant inference costs and latency in production systems while maintaining response quality.

LSTM Drone Scheduling

Nov 2024 · Operations Research · Deep Learning · Optimisation

Long short-term memory network for adaptive drone route scheduling under uncertain demand. Learns operational patterns from historical dispatch data to generate near-optimal schedules, outperforming classical heuristics on dynamic delivery environments.

skills & tools

ML / Deep Learning

PyTorch · scikit-learn · NumPy · SciPy · Keras

Computer Vision

OpenCV · SIFT · FLANN · homography estimation · image rectification

Stochastic & Statistical

Monte Carlo simulation · copula modelling · extreme value theory · time-series analysis

Optimisation

Mixed-integer programs · MAGSAC++ · metaheuristics · operations research

Languages

Python · MATLAB · C++ · bash

Tools

Git · Linux · LaTeX