Projects
If I can't measure that it works, I don't ship it. The rule holds for everything here: engineering systems, peer-reviewed research and practice labs. Filter by type, or browse them all, newest first.
-
Archetype Tracking
Reports 2 Modern archetypes fortnightly for a team meeting. It measures what pilots registered and never performance, and freezes every reading it has reported.
PythonDuckDBmatplotlibuvpytest
-
Deck Optimisation Engine
Mines Magic Online decklists to tune one Modern deck. Its own audit showed the data measures adoption, not performance, and the engine says so.
PythonDuckDBuvpytest
-
7PH Graph: A Knowledge Graph of a Competitive Metagame
A live knowledge graph of a competitive Magic format. It shows which cards travel together and refuses to chart weak evidence.
PythonCypherGradioPlotly
-
Digital Twin
An agent answers questions about my work. A second model checks every answer before you see it. Accuracy scores 4.56 of 5 on 149 questions.
PythonChromaDBLiteLLMGradio
-
AI-JIE: Extraction and Evaluation Pipeline
The extraction layer of the Job Intelligence Engine tells a genuine requirement from a nice-to-have. Human review scored the final prompt at 4.11 of 5.
PythonOpenAI APIPydanticasyncio
-
LLM Engineering Lab
A lab of 11 projects across retrieval, fine-tuning and agents. The flagship prices Amazon products from their descriptions to within $29.95 on average.
PythonQLoRAChromaDBModal
-
Job Intelligence Engine
Sorts 6,162 postings into 2 shortlists, target now or worth a stretch, and names the skill to learn next.
Pythonscikit-learnSBERTXGBoostStreamlit
-
MLB Analytics with SQL
Reusable SQL over 150 years of baseball data answers 4 questions. One finding: the average debutant has grown more than 5 inches taller.
PostgreSQLSQLPython
-
Python Labs
Three lab collections cover the fundamentals: object-oriented Python, data analysis and machine learning. The machine learning collection runs from regression to neural networks.
Pythonpandasscikit-learn
-
Seasonal altitudinal migration in rainforest birds
Rainforest birds move uphill each summer and back down each winter. Sixteen years of counts make it the first system-wide measure of this movement.
RJAGSN-mixture models
-
Forest gap effects on tropical birds
Forest gaps changed which bird species lived where, while total numbers stayed level. One specialist, the Hill Blue Flycatcher, increased with gap size.
RGLMs
-
Physiological stress and rainforest mammal declines
A model chain from microclimate to physiology to demography showed heat stress and foraging limits driving rainforest possum declines, species by species.
RJAGSBiophysical models
-
Climate, foliage chemistry and herbivory
Climate and geology set the stage for how much insects eat rainforest leaves. Single soil nutrients, the expected drivers, predicted little after accounting for geology.
RJAGSHierarchical models
-
Climatic drivers of rainforest bird change
Warming and shifting rainfall drove opposite trends in lowland and upland birds across 47 species. Cyclones and droughts had only marginal effects.
RJAGSRemote sensing
-
Ringtail possum viability forecast
A model on 30 years of surveys forecasts possum collapse by 2050, heatwaves doing most of the damage. The forecast fed a national protection nomination.
RJAGSForecasting
-
Community reshuffling under elevational shifts
A simulation of 7,613 wildlife communities predicts mass local extinctions as warming pushes mountain-top species out of habitat. Each species climbs at its own speed.
RSpatial forecasting
-
Rainforest bird declines
Seventeen years of monitoring showed upland rainforest birds nearly halving. The evidence supported protection nominations for 14 species.
RGLMsShiny
-
Predicting abundance from environmental suitability
Maps of where a species can live predicted about half the variation in local numbers. The models faced 50 species and places they never saw.
REnsemble MLMaxEnt