Work

three research platforms, and the field-to-software work between them

Most of my work follows one pattern: take a piece of research software that would otherwise die as an un-runnable desktop binary, re-implement it in pure Python, pin the re-implementation to the original's output with a regression test, and deploy it as a documented, openly-licensed web service that someone else can actually run. It's a deceptively boring-sounding goal — make the tool outlast the paper — but it's most of what separates research code that gets cited and reused from research code that rots. The three research platforms below started as C++/Qt desktop tools. Three research platforms built that way are live now — alongside current work under my PFP Works banner, and earlier health-tech work for the UK NHS.

Landuse Analyst

la.arkygeek.com — modelling what a region's land and water could support, through time.

LA answers a specific question for a specific region and period: given a reconstruction of rainfall, hydrology and terrain, what range of agropastoral activity could the land-and-water budget sustain? It began as the C++/Qt tool behind my PhD (the agricultural catchment of an Early Bronze Age settlement on the Jordan Rift) and is now a cloud platform plus a QGIS plugin.

• Architecture. A FastAPI orchestration tier; an Astro + Svelte + MapLibre front-end; a separate, off-host compute service for the heavy pipelines, reached over signed service-to-service calls; the whole thing presented as one host through a thin edge worker, with a documented deploy pipeline.
• Engines. Walking-cost catchments (Tobler's hiking function + a binary-search threshold sweep over per-crop suitability rasters), palaeo-hydrology / flow accumulation, and an anisotropic wildfire-spread model.
• The catalogue. The global landscape partitioned into prehistoric land-use regions, each carrying a transparently scored stack of around ninety ingestible datasets — so a pipeline built in one region can be ported to another with a known data footing.
• Reproducible. A pure-Python re-implementation that reproduces the original thesis runs.

DataRanking

dataranking.arkygeek.com — grading the quality of datasets used to calibrate agro-ecosystem models.

Drop in a dataset and DataRanking returns an overall Platinum / Gold / Silver / Copper tier plus per-block tiers across the framework's eight categories, with a radar chart of where the data is strong and where it's thin. It's a web port of the Kersebaum et al. 2015 framework — a paper I co-authored — re-implemented from the original 2014 C++/Qt desktop tool.

• Reproducible by construction. The port is pinned byte-for-byte to the original Windows binary's verdict via a regression test against Kersebaum's own Müncheberg crop-rotation example. This is the single detail I'd point a reviewer to first: it's the difference between "a rewrite that's probably faithful" and "a rewrite that is provably faithful."
• Interoperable. Accepts raw JSON and AgMIP / ICASA .aceb / ACE-JSON files; ships versioned weight sets (the binary's behaviour and the published paper's intent, documented where they differ); exposes an OpenAPI schema and /docs.

Agricultural Model Adapter

mad.arkygeek.com — a model-suitability and configuration translation companion.

The Agricultural Model Adapter (A MAD Toolbelt) is a zero-dependency, local-first browser platform designed to bridge crop, livestock, and farm-economic models. Originally built as a Qt/C++ desktop utility under the European FACCE MACSUR network, this web companion scales the original dataset quality ranking to support multi-model ensembles, FAIR compliance matrices, and direct configuration exporters.

• Multi-Discipline Mappings. Supports Crop (AgMIP/ICASA standard vocabularies), Livestock (LiveM pasture and herd dynamics), and Economic (TradeM adoption rates, regional returns) blocks.
• Cross-Model Linkages. Instantly calculates suitability percentages against major biophysical simulations (HERMES, DSSAT, APSIM, WOFOST, MONICA, AquaCrop, STICS) and warns if critical cross-discipline dependencies are unchecked.
• DSSAT & APSIM Exporters. Automatically translates and formats AgMIP Crop Experiment (ACE) JSON data to space-aligned DSSAT .SOL and APSIM JSON configuration files directly in the browser.

Pyroscape

pyroscape.ca — a cloud-native wildland fire simulation, data adapter, and emergency logistics engine.

Pyroscape is a high-performance, cloud-native wildfire simulation, model translation, and real-time emergency logistics engine. Designed to support emergency response and historical validation, it couples spatial data-ranking engines with stateless model translation APIs to ingest disparate global weather, topography, and fuel inventories, piping them directly into legacy or modern coupled fire–atmosphere engines.

• Real-time Emergency Logistics. An evacuation routing framework coupling Dijkstra pathfinding with a 300-second physics safety buffer, Bureau of Public Roads (BPR) traffic queuing, and Haversine geospatial auto-matching for logistics.
• Secure Multi-modal Ingestion. A resilient data ingestion architecture with deterministic fallbacks from ArcGIS REST APIs to raw XML RSS scraping, secured by FOIPPA-compliant Fernet symmetric encryption for demographic payloads.
• The Pyro-MAD Translator. A stateless translation layer and API that dynamically harmonises Canadian FBP and US Scott & Burgan fuel classifications in-memory via high-speed GDAL and Rasterio virtual file streams, reducing model-initialisation data-wrangling from hours to under sixty seconds.
• STAC Cloud-Cover Verification. An automated, pre-download verification workflow that combines SpatioTemporal Asset Catalog metadata queries with daily NASA GIBS true-colour imagery timelines to identify the first clear-sky, Normalized Burn Ratio (NBR)-capable Sentinel-2 pass over active megafires, saving over 98% in bandwidth.
• Dynamic Scoring. A four-axis Multi-Criteria Decision Analysis (MCDA) ranking framework that dynamically weights environmental datasets based on operational intent (active incident forecasting vs. historical benchmarking).

How they connect

These aren't unrelated side projects. LA's own method for scoring palaeoenvironmental datasets is a direct extension of the DataRanking framework into a new domain — one platform is, in a real sense, the methodological parent of a subsystem in the other. The through-line across all of it is the same: rigorous, reproducible, reusable research software for the archaeological and agro-environmental sciences.

PFP Works

my software & food-production venture.

My current work lives under PFP Works — the name I build software platforms and sustainable food-production ventures under. Two of them: Prairie Fresh Produce, a farm-to-table aquaponics venture, and Prairieline, a multi-tenant retail platform. It's where I do current full-stack product work end to end — architecture, development, deployment and operations.

Earlier work: OpenPCI

health-tech — clinician–patient communication for the UK NHS.

A full-stack Laravel + Vue web application for communication between healthcare professionals and patients, with a simple, robust scheme for protecting patient information. It grew out of a research project at Edge Hill University carried out with NHS clinicians, and the prototype was approved for use at Aintree University Hospital — a notably strict NHS Trust for information governance. Different domain and stack from the platforms above (PHP rather than Python), which is exactly why it's here: the instinct to build durable, trustworthy software travels beyond archaeology into a regulated, high-stakes environment. The repository is private (it's medical software); a code walkthrough is available on request.

Research →  ·  CV →  ·  GitHub →