Professional Background
- Core Focus: Computer vision, Earth observation, and scalable geospatial models.
- Cefas: Engineered end-to-end MLOps pipelines for terabyte/petabyte-scale marine datasets. Overhauled the UK's Benthic AI litter project (presented at ICES), and implemented RAG, NLP, and DAG models for causal analysis.
- British Airways: Served as a Data Engineer Consultant, developing production flight analysis code.
Education & Philosophy
- Degrees: MSc in Space Engineering & BEng in Aerospace Engineering.
- Foundation: Technical expertise built on satellite radar data and UAV aerodynamics.
- Optimization: Focuses on memory utilization and garbage collection in Python pipelines. Builds cost-effective, hardware-agnostic solutions (e.g., Vulkan API, OpenCL) to bypass strict reliance on Nvidia GPUs.
Technical Projects Portfolio
Species Data Pipeline & RAG System
Hybrid codebase integrating marine species taxonomy pipelines with an end-to-end Azure AI RAG ecosystem. Features a Flask REST proxy, Tailwind frontend, and multi-agent orchestration.
Azure AI, Docling, Semantic KernelCefasRAG
Python/Azure app for fisheries advice combining RAG, species taxonomic enrichment, live web automation, and cloud management via a desktop GUI to query ICES PDFs.
GPT-4o, Selenium, Azure AI SearchCloud Development Tools v2.0.0
Desktop app providing a unified GUI for managing Azure Cloud resources and GitHub repositories, built on a strict MVC pattern with async background execution.
tkinter/ttk, Azure-SDK, PyGithubCausal Inference Framework
Implements modern causal machine learning algorithms in PyTorch/Python, including CausalBertWrapper bug fixes for ATE estimation and causal discovery.
PyTorch, DoWhy, EconML, Causal-learnBiological Shape Analysis
Modular pipeline for automated image segmentation, Elliptic Fourier shape analysis, and Bayesian statistical modeling of fish otoliths with GPU-accelerated Fourier analysis.
CuPy, TensorFlow Probability, PlotlyDrone/UAV Annotation Pipeline
Automated annotation counting in UAV datasets, combining ODM log parsing, feature matching, and graph theory using transitive reduction to prevent double-counting.
OpenCV, NetworkX, Segment Treeshexcover
Geospatial pipeline converting OSM regions and point clusters into uniform H3 hexagon covers, evaluating spatial bias using pure-NumPy vectorized algorithms.
NumPy, H3 indexing, Nesterov 2D GJKPlum Image Tools (vision-toolkit)
Comprehensive computer vision monorepo tailored for marine science, unifying COCO manipulation, autoencoder clustering, and geometric testing.
PyTorch, NetworkX, OpenCVvideo-zarr & cctv_zarr
Utilities converting large video files into OME-Zarr cloud-native storage. Includes optical-flow motion tracking and selective chunk-level cloud querying without downloading full videos.
zarr v3, opencv-python, multiprocessingCloud_Dashcam_POC
Extraction pipeline for dashcams combining computer vision with GPX telemetry for coverage chunking, quality gating, and reflectance reconstruction.
Python, GPXpy, OpenCVCompressCompare
C++17 desktop app designed to visually analyze loss from re-encoding via per-pixel ΔE2000 heat maps. High-speed performance on legacy hardware via CPU/GPU parity.
C++, OpenCL, OpenCV, Dear ImGuiAberTech
A suite of three fully functional, commercially viable applications unified through a central orchestration layer to assist mountain search and rescue missions and save lives.
Orchestration, Systems ArchitectureFeatured Writing & Insights
Advanced Computer Vision
Video Frame Compression for AI: Pipeline for vehicle cameras using physics-based exposure gating and coverage-aware chunking.
How I Ditched IOUs: Engineering a custom sweep-line algorithm using segment trees.
The SAHI Polygon Splitter: COCO slicing pipeline that cuts large annotated images into overlapping tiles.
Philosophy & Cloud Architecture
The Inverted Problem: Argues that bounding scope and scaling problems to physical reality prevents severe system failures.
A Lesson in Humanity in AI: Reflections on building resilient technology shaped by solo exploration.
AberTech - Search and Rescue: Proposal for an AI framework combining UAV imagery, ESA data, and dashcams.