boot sequence complete

Twenty-plus years at the seam between art and engineering.

I quit medical school to become a starving artist, and nearly starved to death. So I pivoted. What started as freelance design work became two decades of building products, leading the teams that grew them, and now designing agentic AI systems that ship to production, not just to a deck.

$250K

saved annually, Asurion AI platform

75%

vendor dependency cut

enterprise products, grown from the first apps I built

view the work download resume

brynn_hart / portrait.jpg / 2026

whoami

Clarity from chaos, on purpose.

My background in medicine gives me an edge in healthcare UX, but I thrive anywhere that needs structure pulled out of a mess. I call my design style cozy minimalism: warm, clean, human-centered, with a phosphor-terminal streak I never quite grew out of.

AI has been a throughline in my work since my very first design job, building the application layer around an evolutionary-algorithm engine for Fortune 500 concept testing, long before "AI-native" was a category anyone hired for. For the last three years that thread has become the center of my practice rather than the edge of it, not a feature I bolt onto a project, but core material I design and build with directly. Human-AI interaction, designing when a system should act on its own, when it should ask, and when it should defer entirely to a person, is the specific problem I care about most. I architect agentic systems that generate, evaluate their own output, and are designed to escalate to a human when they should. I write the prompts and the pipeline code. I know the difference between a demo that impresses a room and a system that survives contact with production.

My path has been a steady widening of scope: building the application layer around one algorithm, then a suite of clinical apps and the team that grew it, then an organization's shared AI foundation, and now my own agentic systems. Where I'm headed is executive AI leadership. I want to be a Chief AI Officer: the person who decides where an organization uses AI, how much autonomy it gets, and how it earns the trust of the people who rely on it.

I'm at my best when someone hands me a whiteboard sketch of a hard problem and needs it turned into something a real team relies on by Monday morning.

// stack

Adobe Creative Cloud, Firefly pipeline architecture
Claude API, OpenAI API, Google Gemini, Azure AI, MCP
Figma: systems, tokens, plugin architecture
React, Angular, Vue, Node.js, Python
Claude Code: daily driver
D3, Tableau: data visualization

selected work

Six systems, one throughline: make the complex thing legible.

Case studies below, ordered from newest to earliest. Read from the bottom up and you can watch the scope grow, from the tooling around one algorithm, to a suite and the team behind it, to an organization's AI foundation, to agentic systems of my own.

~/work/agentic-asset-generation · 2024–present

Agentic system architecture diagram (Watcher → Generator → Evaluator → Alert)

Python · Google Gemini · CLIP

Building an Agentic AI System for Asset Generation

Enterprise creative teams needed brand-safe assets at a volume no human review process could keep up with. I designed a four-stage pipeline to generate, evaluate, and deliver creative assets, without losing the brand-safety net a human reviewer provides, and I'm building it out stage by stage on Google Gemini rather than shipping a black box all at once.

Architected a four-stage workflow (Watcher → Generator → Evaluator → Alert), with the human-in-the-loop decision, when to act, ask, or defer, built into the design from the start
Built and shipped the Generator on Google Gemini, producing multi-ratio asset variants from a campaign brief, with constraint-aware text layout and hero-asset reuse
Built and shipped the Evaluator: CLIP concept-match scoring, perceptual-hash duplicate detection, and compliance checks on every generated creative, with a per-creative run report
Watcher (trigger monitoring) and Alert (escalation to a person) are the remaining stages, in active development now

Result: two of four stages live and on GitHub today. Closing the loop on the full autonomous pipeline is the current build in progress.

agentic AIGoogle GeminiCLIPcompliance design

↳ github: Generator + Evaluator ↳ CLI demo

~/work/genai-tool-suite · asurion, 2023–2024

Custom GenAI tool suite dashboard, hero shot

Asurion · Figma · ChatGPT API · Azure AI · DALL·E 3 · AWS

Building the Org's Central Brain for GenAI

Every department, and every AI tool built on top of our stack, was answering brand, policy, and product questions a little differently, because nothing forced consistency. Creative, support, and strategy teams were also stuck in manual content workflows with no shared system for brand-safe generation. I owned this product end to end, from requirements and specification through frontend, integrations, prompt architecture and rollout, building an internal suite of branded GenAI tools anchored by a RAG pipeline that became the closest thing the organization had to a second brain: one grounded, curatable source of truth that both the creative-workflow AI and the customer, vendor and partner chatbot drew from.

Ran user discovery to prioritize real workflows over hypothetical ones
Built a RAG pipeline on Azure AI grounded in a curated SharePoint knowledge base, refreshed nightly, covering brand voice, legal and compliance constraints, product history, and policy, so the AI tools built on top of it gave the same answer instead of drifting apart
Built prompt frameworks giving non-technical teams control over tone, structure, and intent, without giving up the shared ground truth underneath

Result: rolled out across 3 departments with training, demos and documentation, saving $250K annually. A separate production image-generation pipeline cut external vendor dependency by 75%.

RAGGenAIAzure AIknowledge governancecross-department adoption

~/work/health-analytics-dashboard

Miro · Figma · D3 · Tableau · Adobe Creative Suite

Healthcare Quality Analytics Dashboard

Executive and care-coordinator audiences needed to make sense of dense quality metrics and discharge patterns without a data-science background standing between them and a decision.

Ran persona-based journey mapping and stakeholder interviews
Surfaced pain points in data comprehension through Miro workshops
Built D3 visualizations prioritizing clean hierarchy and accessibility

Result: faster, clearer decision-making from the C-suite down to care coordinators.

D3.jsTableaudata visualizationhealthcare

~/work/patient-transition-platform · navihealth, 2013–2022

Miro · Sketch · Angular 12 · Bootstrap 4 · React Native

Hospital Patient Transition Platform

Discharge handoffs were failing on poor documentation and inaccessible plans, raising real readmission risk. I redesigned the workflow from the ground up and built the original apps myself. Then I led an interdisciplinary team of engineers and designers who grew them into twelve products on one shared design system, with the contribution model and adoption standards that let other teams build on it without me in the room.

Designed and built the original apps, then led the engineers and designers who expanded them into a 12-product suite
Conducted field research directly with discharge nurses and case managers
Pinpointed handoff errors caused by documentation gaps
Built situationally-aware UIs tuned for fast data entry on mobile and web
Owned the roadmap and governance of the enterprise design system under the whole suite, driving adoption across engineering and product teams and improving engineering velocity by 40%

Result: faster, safer handoffs with consistent experience across every surface, and a design system that kept running across 12 products years after I left.

design systemsgovernanceAngularReact Nativefield research

↳ full design system

~/work/evolutionary-concept-testing · affinnova

Self-serve variant-building editor, before and after

Affinnova (now Nielsen Optimizer) · Whiteboard · InDesign · Backbone.js · jQuery · Illustrator

Building the App Layer Around an Evolutionary Algorithm

Long before generative AI was a category anyone hired for, Affinnova was running Fortune 500 concept testing on a patented evolutionary algorithm. I didn't build the algorithm itself. I built everything around it: the tooling that assembled visual stimuli, fed real consumer preference back into the model generation over generation, and eventually let clients build their own test variants without waiting on our design team.

Built tooling that auto-assembled taglines, imagery, and packaging concepts into visual stimuli for large-scale online consumer testing
Fed preference signals back into the algorithm each generation, surfacing the strongest-performing combinations to Fortune 500 clients
Expanded the methodology over time into richer signal capture: click-based preference mapping, attribute-level rating, and eye-tracking
Designed and shipped a self-serve, Photoshop-style variant-building app that let clients construct their own test stimuli directly, my last and biggest project there

Result: the technology is now Nielsen Optimizer, still running concept tests for some of the world's largest brands.

evolutionary algorithmsML application designNielsen Optimizergenerative testing

↳ App demo

~/work/matchmaking-app · personal

Affinity Designer · Angular · Ionic · Affinity Photo

Matchmaking Mobile App

Swipe-based dating apps were producing real fatigue and shallow matching. I designed a new take centered on personality over photos, with full-stack ownership from research through rollout.

Researched swipe fatigue to surface real emotional pain points
Designed a question-first matching flow to promote deeper discovery
Built the entire mobile experience with calming visuals and soft haptics

Result: a more human approach to dating, owned end to end from concept to rollout.

Ionicmobile UXpersonal project

~/side-projects

Off the clock, same throughline.

The art-and-engineering fusion isn't a work persona. It's just how I build things, all the time.

Eurorack instruments

Hand-soldered synth and drum modules on the Daisy Seed platform: 3D-printed panels, hardware I design and build myself.

Dead Internet Society BBS

Sysop of a full phosphor-CRT-aesthetic BBS: Node.js, WebSockets, a homegrown chrome economy, and a roguelite door game.

Goth Girls vs The Goo

A cybergoth platformer IP: worldbuilding, pixel art, and game design across a Steam release and door-game spinoffs.

~/contact

Say hello, or send a brief.

Open to senior AI product, platform and leadership roles, on the path to Chief AI Officer.

brynnhart@gmail.com download resume linkedin