I'm interested in local models, agent workflows, research tools, public-service decision support, and web applications built around real human decisions, measurable signals, and practical use.
My interests include running large language models locally on my own hardware, and using AI agents for practical work: software development, business and marketing strategy, research-library search, literature corpus building, data handling, and structured analysis. I test models myself rather than taking benchmark claims at face value, because the useful question is not whether a model sounds impressive, but whether it behaves reliably inside a real workflow.
My wider technical interest is in frequencies and signals of all types: biological, acoustic, behavioural, digital, electrical, and mechanical. The most interesting work sits where those signals intersect, where raw activity becomes measurement, pattern, decision, interface, or intervention.
The projects below sit in that territory. Some are public-facing tools for families, travellers, riders, and local residents. Others are decision-support systems for organisations where clarity and evidence matter, including recruitment, public services, policing, defence, emergency services, and other high-consequence environments.
Why do new ideas, and new machines, only become usable once they have a name? From Aranda skin names to the Pythagorean catechism ("What is the wisest thing? Number. What is next in wisdom? That which gives names to things.") to the personas, nicknames and namespaces we give AI systems: naming as the second tier of wisdom.
Models, datasets and other releases from the lab, published openly on Hugging Face so anyone can download, run, test or build on them.
Hands-on benchmarks of local language models on NVIDIA DGX Spark hardware. Every model is tuned to its own best configuration and measured on agent tasks, tool use, speed, speculative decoding, handwriting OCR, and stability under real workloads. The interactive scorecard lets you rank the models by whichever measures matter to you, and every cell shows the raw measurement and the exact configuration behind it.
My scoping review on EEG and ECG in music listening is now something you can question directly. Built with the Paper2Agent method of Miao, Davis, Zhang, Pritchard and Zou (2026), Reimagining research papers as interactive and reliable AI agents, Nature, it pairs the published paper with a chat box that answers only from the paper's own text and cites the section it came from, a screener that applies the review's eligibility rules to a candidate study, and a self-scoring checklist built from the same charted data. The paper, its appendices and the tool are all open source.
A research publishing project about music, psychological processes and bodily responses. The public website explains the scope, author guidance and approach to transparent AI assessment, with a final human decision before a paper is published.
My music psychology and sound strategy consultancy for spaces, brands and experiences. Sonic Reach connects research with practical decisions about conversation, comfort, music and atmosphere, with interactive acoustic demonstrations on the public website.
A plain-language aged-care information site for Australian families who need to understand care options, eligibility, costs, decisions, and next steps. It is designed to reduce confusion at a stressful point in life, turning government language and fragmented advice into a clearer path forward.
An Australian labour-market guide that brings together official jobs, skills, wages, economy and living-cost data. Explore roles and places with the source, reference period and limits of each measure kept visible.
A working structured preparation portal for candidates preparing for rigorous entry and selection requirements. It provides focused pathways including PolicePath for New South Wales Police Force applicants, ArmyPath for Australian Army candidates, and ClearancePath for Australian Navy clearance diver preparation. It helps candidates understand role requirements, organise evidence, prepare stronger interview responses, make better personal decisions about fit and readiness, and arrive selection ready.
A digital companion for Berrima Diesel customers and touring travellers. It brings diesel knowledge, trip-relevant support, workshop information, and practical service guidance into a more useful interface for 4WD owners, caravan travellers, remote-area drivers, and people who need their vehicle to get them home.
A browser-based music perception study built around my interest in rhythm, frequency, timing, taste, and preference. It lets people try a short interactive task, choose a comfortable tempo, and report how simple rhythms group perceptually. The study turns subjective experience into structured data without needing lab hardware.
A template for building a rated literature corpus.
A simple civic utility for remembering the right bin on the right night. It is useful for residents, new renters, share houses, property managers, and real estate agents who want to give new tenants a practical local tool instead of another forgotten fridge magnet.
A motorcycle rally and event aggregator built from my love of motorcycling, touring, travel, and the brotherhood of riders. It is intended as something useful to give back to the riding community: one place to discover rallies, plan trips, and keep event information from disappearing into scattered posts and pages.
An Australian map of public payphones, free Wi-Fi hotspots, and Australia Post postboxes. It shows where these services can be found and helps travellers, people without reliable mobile data, emergency planning, and local service awareness.
A machine-learning forecasting experiment for Bitcoin price movement. It is not financial advice; it is a practical test bed for time-series modelling, noisy market signals, feature selection, model behaviour, and the limits of prediction when human behaviour, liquidity, news, and technical data collide.
Wi-Fi Radar turns nearby wireless signal patterns into an operational picture of how spaces are being used in real time. It helps people quickly spot coverage gaps, interference, crowd movement trends, and risky blind spots when reliability, resilience, or situational awareness matters. What makes it different is the combination of radar-style event flow, spatial pattern analysis, and practical decision support in one compact view for real environments.
An Android HRV biofeedback and cardiac monitoring app built around the Polar H10 chest strap. It combines real-time HRV, resonance-frequency breathing, heartbeat-synchronised breathing cues, ECG-derived signal features, and environmental-physiological correlations into a practical personal monitoring platform. It is developed and usable, but not publicly distributed.
A working Jellyfin-first Android app that measures heart-rate variability while music is playing, then scores songs by autonomic response. It connects to a Bluetooth heart-rate chest strap, processes HRV metrics, tracks sessions, exports data, ranks tracks by coherence-style response, and can create Jellyfin playlists from the songs that produce the strongest physiological effects. Public release is not available yet.
A private working rugby league prediction app for my own use. It collects fixtures, odds, and team data, runs a machine-learning prediction pipeline, and compares predicted outcomes with actual results so I can test how well the model behaves across real rounds. It is developed and operational, but it is not public release software and has no public link.
A private working scouting and selection app for touch football coaches and selectors. It manages player profiles, multiple scouting reports, attribute ratings, coaching brief fit, player combinations, squad-building decisions, and selection evidence in one place. The same structure can be adapted for other sports that need structured player assessment and squad selection. Please contact me for more information or to request a demo.
A handheld bird-call translation concept that maps spoken human intent, such as danger, food, territory, or friend, into species-relevant Australian bird calls. The project combines edge AI, environmental audio, portable hardware, and bio-acoustic signalling.
A handheld underwater sonar idea inspired by compact motion-tracker interfaces. The goal is a practical diver-facing sensing tool for presence, range, and underwater awareness rather than a large boat-mounted sonar system.
An electrical impedance tomography project for detecting and tracking people in water by measuring conductivity changes. It explores low-cost electrode arrays, ESP32-class hardware, and non-camera sensing for pools, ports, security, and underwater safety.
A broader research collection on non-optical sensing through water, including acoustic methods, electrical impedance tomography, magnetic approaches, and the physics that make ordinary Wi-Fi sensing fail underwater.
An exploratory sensing line looking at whether wireless channel changes can reveal biological movement, presence, breathing, heart-rate-related activity, or other subtle human signals without wearable sensors.
The common pattern is practical applied software: research tools that collect cleaner data, public utilities that remove repeated decisions, and small domain-specific apps that make information easier to act on.