AI Systems
Designing an AI system is still design.
You study the people, map the flow, prototype, test, and refine. The material is different. The practice is the same.
My specialty is AI for work where the numbers matter: finance, insurance, and risk. Automation that a skeptical underwriter would actually trust.
↑ the idea is the easy part
Experience
I design corporate-level multi-agent AI workflows for regulated industries.
Case study 01 · Flagship
PerilAI
LiveAgent-readable
An agent-readable ledger of real AI-liability matters, each filed by the insurance line it hits and carrying its own sources, served through an API built to drop into agentic workflows.
See the full case study →← the landing pageCase study 02
Artie Studio
Workflow designHuman-in-the-loopSelf-improving loopsApp design
The problemA researched video essay takes dozens of steps across research, script, storyboard, and image generation, spread over too many tools.
What I designedOne production app that runs the whole pipeline: research, script, beat sheet, image prompts, storyboard, and sequence, with a checkpoint wherever my judgment matters.
How it learnsEvery script edit and image review feeds back into the system, so its drafts get closer to my voice and my visual style over time.
narration on the left, frames on the rightFig. 02 · Storyboard view: narration lines paired with illustrated frames.
Case study 03
Skill Library & Atlas
Systems thinkingProcess design
The problemAI assistants forget how you like things done.
What I designedA library of 80+ custom AI skills. Each one encodes a repeatable workflow, like research, fact-checking, or visual direction. A searchable Atlas sorts them automatically, so the whole system stays easy to find and easy to improve.