Clients, an engagement, and the lab.
Kept apart on purpose. What a client paid for and what I built to prove a technique are different claims, and mixing them would flatter both.
[×] REEL 01 · CLIENTS · 2 ENTRIESIN THE GATE
Built and delivered for clients. Some of it is public; some has not been released.
Projects are described at concept level. Anything unreleased or internal to a client stays that way.
State Auto Glass
A mobile windshield repair business has served the mid-Willamette Valley since 1961. I built its live site, then studied the market and competitors, and wrote a growth plan.
PROBLEM
A sixty-year-old local business needed to be found by the people already looking for it, across a dozen towns — and to turn that traffic into quote requests.
RESULT
A fast, static Next.js site: quote form, interactive coverage map, service index, field report, and a dedicated page for each town served.
Elite Gamers
A full Arabic-first gaming publication — news, reviews, video, hardware, guides and esports — on a bespoke WordPress theme: nine custom post types, a weighted review-scoring engine, twelve hand-written Schema.org types and a 147-token design system, with no page builder, no SEO plugin and no build step.
PROBLEM
Arabic-speaking gamers read their news through interfaces built for English first, with right-to-left bolted on afterwards — and no theme you can buy carries a publication's nine content types or a review score its editors can trust.
RESULT
A theme written end to end: 9 post types, 4 taxonomies, a weighted multi-axis rating engine with a single source of truth, a ⌘K command palette with Arabic orthographic search folding, and a live AR / FR / EN toggle. Around 1.7 MB cut from the hero payload, and audited after completion across 11 domains — 96 findings confirmed and resolved, zero critical.
An engagement leading the company's IT and development function.
Described at concept level: what each build was, and nothing about the company behind it.
Revolution Conseil
Senior Developer — Head of IT & DevelopmentApril – June 2026
Ran the IT and development department — its day-to-day administration and infrastructure, plus the work below. I owned each of these from brief to hand-over: design, build, test.
- ERP desktop application — Windows and macOS.
- Offline archive browser for a theatre — a premium local web app for browsing the full archive on a large tablet, no connection required.
- Tourism website — the largest build of the engagement, finished and awaiting release.
- Real estate website — in progress when the engagement ended.
- mondoenergie.com ↗ — mobile layout corrections. Built by another member of the team, not by me.
Some of this is live, some runs inside the company, and the rest is finished and waiting to be released by the company.
Built for my own use, not for a client. This is where a technique gets proven before it goes near someone's business.
Vision Gate
An access-control and attendance system that recognises faces at a door and keeps the record — even when several people walk through together.
PROBLEM
Badge and sign-in sheets drift from reality. Who actually entered, and when, is a question most small businesses cannot answer — and a single camera watching one face at a time misses a group arriving together.
RESULT
A working system, tested with groups of up to five people walking through the gate at once: it separates and identifies each person individually, and logs entry and exit for every one of them to build an attendance record. Built and exercised; what remains before release is a security hardening pass and licensing.
MT5 Trading Terminal
A desktop trading terminal wired into MT5 and a live news feed, with a pure-mathematics analysis core underneath it.
PROBLEM
Tracking several markets by hand is slow, and a scattered view produces bad decisions. The macroeconomic data that would fix that sits behind paid feeds priced for institutions, not for one person's tooling.
RESULT
A terminal carrying a full market dashboard, an embedded TradingView chart, execution controls, and demo-account support alongside live accounts. The economic calendar and macro data arrive through a custom MT5 expert advisor I wrote for the purpose, so the terminal imports them directly instead of paying for a feed. Underneath sits the calculation core: multiple gold and crypto strategies across several timeframes, with market-structure detection choosing which strategy suits current conditions. Local and hosted language models — Ollama, LM Studio and vMLX among them — turn the result into clean JSON signals with the reasoning attached, rendered in plain language: buy, sell, long, short, or close one, several or all positions. Around 90% built; what remains is testing.
I’m not a professional trader or financial advisor. These are tools I built to support my own research and decisions. It performed well on demo accounts and it also took losses — trading is not something to hand over to an automated system alone.
OpenClaw macOS Bridge
A bridge giving an AI agent operator-level access to macOS — mouse, keyboard, and the screen itself.
PROBLEM
An agent confined to a scripting panel can only reach what that panel exposes. Most real desktop work happens outside it.
RESULT
Mouse and keyboard control, screen reading through OCR, and a custom action layer on top — enough to drive an application like Blender directly rather than through its scripting interface, and to carry out browser-based work end to end. Built while the platform had no agent layer of its own.
Sophia AI Agent
An all-in-one assistant — chat, voice, screen reading and desktop control — built well before this class of agent was common.
PROBLEM
An assistant is only useful if it meets you where you are: by voice, by text, and on whatever is already on the screen. Doing that means leaving the chat window and touching the machine itself.
RESULT
Text and voice conversation through TTS and STT, OCR to read the screen, mouse and keyboard control through a custom controller, game interaction, and blockchain scanning for smart-money analysis. Model quality was the ceiling, not the design — the models of the day hallucinated often — but the shape of it was right.
OpenClaw Desktop Dashboard
A desktop control surface for OpenClaw — chat, management and control in one window, built before the product shipped a dashboard of its own.
PROBLEM
Driving the system meant moving between separate interfaces, and part of what I wanted to do had no interface at all.
RESULT
One desktop application to chat with the system, manage it and control it, carrying capabilities the product did not ship at the time. Finished and stable.
Pine Script Strategies
TradingView strategies for gold and crypto — analysis, signal generation and technical workflows.
PROBLEM
Testing an idea across instruments and timeframes by eye does not scale past a couple of charts.
RESULT
Strategy scripts covering market analysis, signal generation and the technical trading workflow, written for gold and for crypto.
I’m not a professional trader or financial advisor. These are tools I built to support my own research.
Low-Level Network Research
A self-directed study in binary analysis and low-level networking: reverse engineering a closed-source Windows utility, then rebuilding an equivalent in C++ to understand it.
PROBLEM
Reading about a protocol is not the same as knowing it. I wanted to understand how a network diagnostic tool discovers devices, what it puts on the wire, and why.
RESULT
Binary analysis in Ghidra to recover the program's logic from decompiled C, alongside packet and protocol analysis in Wireshark — ARP behaviour, device mapping and traffic monitoring on a local network. From that investigation I built a proof of concept in C++, working the protocol handling out against observed traffic rather than documentation, and debugging it the same way. An exercise in systems programming, interoperability and cybersecurity fundamentals — educational research, not a product.