Projects
Each case study covers the problem, the architecture, what I was responsible for, what went wrong, how it was tested, and exactly what is not built. The last of those is on the page rather than left out.
No fabricated evidence. There are no invented screenshots, user counts or performance figures anywhere in these case studies. Where a product is live, the live URL is the demonstration. Where it is not, an authored architecture diagram describes the real system instead.
Case studies
Two released desktop applications built on locally hosted AI models, an autonomous agent, a deterministic simulation engine, and the generator that builds this site. Each is described in full, including what is not finished.
AI-assisted animation production platform
A desktop application that turns a written episode into a finished 4K animated video: script, cast, voices, lip-sync, camera, render, score, mix and credits.
Live AI dubbing for anime, running locally
A Windows application that gives every character in a show their own English voice as they speak, with the soundtrack left intact and an AI watch companion that cannot spoil the episode.
Autonomous AI presenter and live-stream host
A desktop application in which an animated AI character hosts a live stream unattended — reading chat, choosing topics, reacting to what is on screen and controlling her own performance, with no operator pressing buttons.
Deterministic social-reasoning engine with an enforced information boundary
A dependency-free simulation of eight characters who hold private knowledge, remember what they are told, track who owes whom, lie, and vote — with a test that proves the AI code cannot read the ground truth it is supposed to deduce.
Zero-dependency static site generator and publishing toolchain
The site you are reading, and the tooling behind it: a generator with no dependencies, a build that refuses to publish broken output, and document generation written down to the file format.
Video analysis, scoring and short-form editing from one upload
Upload one video. The application watches it, listens to it, measures every part of it, tells you where it goes wrong, and cuts the strongest moments into short-form clips — all from that one file, with nothing invented.
Also built
Neither of these is big enough for a full case study, and writing one would stretch it. They are here because they are real and checkable.
Diagnostic tooling · PowerShell, Python
A watchdog, a stress harness and a crash reporter written to trace a repeating display-driver failure on a render workstation. The watchdog samples GPU state to a timestamped log, the stress tool reproduces the condition on demand, and the reporter collects the system state at the moment of a hang.
Why it is here It found a verified root cause instead of the usual cycle of reinstalling drivers and hoping. This is the most directly relevant thing here for IT support work: a real intermittent hardware fault, instrumented rather than guessed at.
Used, and it did its job
Web application · Node.js
A shop that holds no stock. Product data and availability come from supplier feeds, a tiered markup is applied, and once an order is paid the admin view states exactly what to order and from which supplier. Includes a PC part-compatibility builder and an automated test suite covering pricing, compatibility, feed parsing and payment-signature verification.
Why it is here The interesting part is the pricing engine: rules that have to hold simultaneously — tiered margin, a minimum profit after transaction fees, tax, and rounding — which is exactly the kind of logic that is quietly wrong until it is tested. So it has tests.
Working prototype, sample data, payments simulated