Shipped with
Three pieces of shipped work, each a different kind of evidence: scale, craft, speed. Every case study opens with the problem, the role, and what actually shipped.
Disney's design management system. Component library, design tokens, governance, used in production by 10+ application teams across the streaming and media division.Why it matters here: Shipped design-system infrastructure used in production by 10+ teams. The same systems-thinking I bring to AI architectures today.
A confidential UX engineering project at Apple, aligned with the Marcom style guide. Dual role across UX/UI design and front-end engineering, held by one person.Why it matters here: Apple-grade integration of design and engineering as a single craft. The working model I now bring to AI systems work. One craft, three specialties, everyone reads everyone else's work.
An MVP scheduling tool for Disney's content release planning. Drag-and-drop calendar grid, dynamic date adjustments, multi-view layouts. Designed and shipped in two sprints.Why it matters here: Speed of execution. Two sprints from "we need this" to a tool teams used in production. The same shipping cadence is what separates working AI systems from working AI demos.
I had the pleasure of working closely with Timurtek on several projects at Disney. He brings a rare combination of design sensibility and engineering precision — the kind of collaborator who closes the loop between concept and shipped product without losing the intent of either. His work elevated every team he touched.
Timurtek is a skilled, and detail oriented UI/UX Engineer who I’ve had the pleasure to work with over 3 years at Disney. His ability to produce pixel perfect user interfaces is rare, especially when combined with Timurtek’s friendly, and positive attitude. As a strong team member, and an expert in UI and UX, Timurtek has helped produce complex component libraries, and features to serve critical business needs.
The AI layer and the design function at an Amazon analytics platform for agencies and brands. The surfaces a customer touches, and the machinery underneath that lets a small team ship them at volume.
The design system and the gates around it, the Report Center 2.0 interface, the marketing site from commit zero, a knowledge base with its own retrieval-backed chat agent, and the org brain that turns every meeting into a searchable record.
The Data Hub interfaces, design first: the production plan gates that work on the designs before a screen gets built.
The marketing site and the knowledge base are public at mixshift.ai. The case study covers the rest, limited to what the repositories show. MixShift’s numbers land when they say so.
One team at a time, the same way whether it is a role or an engagement.
A product team with more workflow than hands. There's an AI roadmap, or the pressure to have one, and nobody whose job it is to turn it into systems that run in production. I take the layer between the model, the interface, and the people who have to use both.
With the one workflow that costs the team the most hours. I model what fixing it is worth before anything larger gets built, and if AI isn't the answer for it, I say so.
Your Slack, your standups, your codebase, your on-call rotation if it matters. I ship inside the stack with the engineers: the pipelines, the eval infrastructure, and the operator-grade interfaces that turn model output into something the team opens every day.
Code, prompts, infrastructure, runbooks, and the habit of running them. I build so the system outlives my hands on it, whether I stay on the team or hand it off.
For a decade I was the designer who could ship and the engineer who could design, at Apple, at Disney, and at a run of startups. The work was design systems: the layer that decides whether a hundred product teams ship coherent software or a hundred slightly different ones.
AI systems fail the same way design systems do: they work in the demo and fall over in operation, because nobody owned the layer between the model and the people who have to use it. That layer is what I now build. The longer version →
Independent products, owned end to end: finding the problem, designing, shipping, operating. The same model I bring to an engagement, with nobody to hand off to but myself.
Decision-intelligence platform for traveling skilled trades workers. Real pay data, contractor safety records, housing intel, and area ratings, built for the 500K+ tradespeople who travel for contracts.The studio venture at full surface area. Marketplace economics, federal-data ETL, content automation, paid SaaS billing, full SEO posture. All shipped solo with Claude as a daily collaborator. The operating model on display.
Read the case studyWhat I’ve learned shipping AI systems that have to survive operation, not just the demo. Essays, not a newsletter; the newsletter is below.
Five agent steps at ninety percent each is fifty nine percent end to end. Agent reliability compounds, so move the checkpoints instead of shopping for a model.
Founder LessonsA file-based content pipeline mapped stage by stage: one plain-text vault, scheduled agent routines, status-driven handoffs, and one human review gate.
Tools & StackWhich MCP servers to connect and which to skip. A verified five-server list, the context cost lens, and one decision rule for when a server earns its slot.
Hiring for a senior or staff role where product thinking meets production systems, or a team that needs this for a season? Thirty minutes, no sales flow.
Not there yet? Read how I got here, or start with the case study that best matches what you’re trying to build.
