Offerings

An agent will get more done for you, and it arrives the way a new employee does — able to work, knowing nothing about your business. It has to be told what the job is, why it matters, and a long list of things never to do. I'm the person who talks to the robot for you: I learn how your business actually runs, then set the agent up with the tools, the skills and the guardrails it needs to be useful here — your invoicing and accounting, your follow-ups, your reporting. Out of the box it runs fine. What it doesn't have is your business.

I build the machinery this runs on and use it every day on my own company's work. lean-ctx is a context runtime I contribute to — the layer deciding what a model sees and when — and APES is the project memory that keeps a dozen agent sessions working from the same truth instead of contradicting each other. Brown Family Sports' software is built that way daily, this site included. The business half isn't new either: I built and grew the online presence for Steph Brown Physiotherapy and Wellness with Kate, where the question was never the technology, it was what actually brings clients through the door.

I come in at the hardest problem you want AI pointed at and say honestly whether it is the tool — often it is not, and that answer alone earns the engagement. Then an audit of how engineering actually ships: where work queues, where it waits, which bottlenecks agents relieve and which ones they only feed. It happens with your teams rather than around them, so the adoption holds after I leave. And what ships gets tied to the outcome it was meant to serve — lines of code never mapped to revenue, and now that agents write them by the thousand, volume means nothing at all.

In Q2 2024, on Sonnet 3.5, I had agents working inside an enterprise system — a distributed microservice estate past 500,000 lines, parts of it opaque even to the teams running it. Out of the box that was not economical, so the research and the engineering came first: a memory system holding what the codebase knew, and a harness around the agents working in it — what each one sees, what it may touch, and what it carries between runs. lean-ctx, the context runtime I contribute to, and APES, the project memory that keeps a dozen sessions working from one truth, are those same ideas built in the open. Two years on, better models have not changed what the work is.

Design and delivery of data-intensive systems: ingestion and transformation pipelines, geospatial routing and mapping engines, and transportation-logistics modeling — turning road-network, terrain, capacity and readiness data into evacuation routing and preparedness intelligence — alongside the dense business logic that sits behind regulated, high-visibility products. Correctness, observability, and uptime are treated as features, not afterthoughts.

On intellimass.ai I own the logistics of evacuation-route condition mapping: a multi-sensor, ruggedized field-deployed device gathers road and terrain conditions from inside the disaster zone and turns them into live, routable evacuation maps — so communities and the teams supporting them know which roads are actually passable when it matters most. I've also shipped engineering inside Telus's eVS (electronic Validation System) ecosystem, where a single defect is visible to an enormous customer base, so the bar is governance, peer review, and zero-drama deploys.

Custom high-performance native applications for the desktop — built directly against the GPU with Rust and gpui for instant, fluid interfaces that simply are not possible in a browser. Hot-reloadable theming, tiling multi-pane layouts, and pixel-level polish throughout.

The Delightful series is the proving ground: Terminal Delight, a GPU-native Linux terminal with hot-reloadable CRT themes and tiling panes, and Markdown Delight, a tiling Markdown editor with Google-Docs-style comment mode. Both are real daily-driver tools shipping today — not demos.

Design systems built on real tokens — color, type, spacing, and motion — so an entire product re-skins from a single source of truth. Whether it is a ground-up system or a redesign of something that has drifted, the result is consistent, themeable, and maintainable: components that compose, themes that swap live, and a visual language the whole team can build on.

This very site is the proof — six complete, switchable design systems (Default, AI, Hacker, Papyrus, Chalk, and a military-grade Tactical theme), all driven by one token layer and swappable from the header in real time. Open any of them below and use the header icons to jump between systems. The Delightful series carries the same idea into native apps with hot-reloadable themes.

Technical SEO and the emerging discipline of GEO — making sure your business shows up not just in Google, but in AI-generated answers. Paired with hands-on online-presence coaching: positioning, content, and the practical business-development work that turns visibility into booked clients.

I built and grew Steph Brown Physiotherapy's online presence from the ground up — a fast, well-structured site engineered to rank locally and convert visitors into booked patients, with the coaching to keep it growing.