Portfolio · Product design & applied AI Available now
Lead Product Designer

Complex systems, made legible.

Twelve years inside mission-critical systems, now designing the AI-native workflows that run them: the pipelines, records, and decisions a business cannot afford to get wrong. Everything below shipped, constraints included.

Position

I work on the parts of a business nobody can see clearly: the pipeline no one can follow end to end, the intake no one trusts, the AI output no one can check. My job is to make plain what is known, what the software guessed, and who signs for the difference.

Selected work Three systems, documented the way they shipped
01

IoT Connectivity

Quote-to-order and CPQ at scale for enterprise IoT: intake, pricing, and contracting on one decision spine.

T‑Mobile for BusinessLead Product Designer · 2024–26
Decision spine Fig. 01.1 · signal → intake → pricing → contract
Sales · opportunity to contract01 / 03 · production
Opportunity page: stage path and links to every child artifact of the deal
Quote to order: product groups forked by fulfilment path
Quote to contract: custom pricing lines, negotiable items, parent contract in the header
The same deal record, three stages down the spine. A configured opportunity becomes a priced quote, then a contract a seller can defend line by line.
Four disconnected tools became one spine. Each stage hands the next a decision that is already made — which is what makes the quote at the end defensible.

IoT connectivity is the cellular service that links a fleet of devices back to the business that operates them.

Sensors, asset trackers, vehicle fleets, payment terminals. An enterprise deal connects thousands of SIMs and eSIMs across whichever network technology fits the workload, so it can’t be quoted from a price list. AI-assisted lead scoring feeds a nine-section intake that drives CPQ, custom and special pricing, and approval.

Four disconnected tools became one accountable surface: graded signal → routed intake → priced configuration → contracted outcome.

Surface
Q2O · CPQ · contracts
Networks
NB‑IoT · LTE‑M · LTE · 5G
Scale
Thousands of SIM / eSIM per deal
Role
Design lead, end to end
02

ElancoGPT

A federated-LLM decision surface with a designed epistemic structure: fact, inference, risk, action.

Elanco Animal HealthSenior Lead Product Designer · 2022–23
Epistemic structure Fig. 02.1 · every response is classified before it is read
ElancoGPT chat interface with a structured answer
ElancoGPT default web surface
ElancoGPT experimental surface
ElancoGPT mobile state board
Chat surface: the model named in the chrome, the answer read as classified parts.
The structure, shipped. Model and risk are named in the chrome; the answer is read as classified parts rather than one undifferentiated paragraph.
Enterprise AI fails on accountability before it fails on accuracy. Classifying the response is what lets a reader act on it — and lets an organisation say who signed.

Enterprise AI fails on accountability long before it fails on accuracy.

Elanco Animal Health is a global animal-health leader of roughly 9,000 people, formed from Eli Lilly’s animal-health division and the Bayer Animal Health merger. ElancoGPT names its model and its risk in the chrome: every response is classified before it is read, so a reader knows what is established, what is inferred, and what carries exposure.

Shipped in a 72-hour window, possible only because the taxonomy and the design system were already load-bearing.

Models
GPT‑3.5 · PaLM 2 · RAG
Structure
Fact · inference · risk · action
Oversight
Human-in-the-loop by default
Role
Design lead · applied AI
03

Elanco Design System

Three-tier tokens, public decision logs, codemod-backed deprecations: the chassis everything else ran on.

Elanco Animal HealthDesign systems lead · 2022–23
System board Fig. 03.1 · three tiers, four specimens, one path from value to component
The system, published
Responsive breakpoint chart
Breakpoints: one grid across every surface.
from semantic
Type ramp
Elanco Design System type ramp
from semantic
Palette · primary + secondary
#0072CE285C #0011492766C #0EC3A43265C #D5C4AF2309C
from primitive
Buttons · states
Button component states
from component
Inputs · date picker
Input and calendar components
from component
Every specimen on the right is produced by the tier wired to it on the left. Change a primitive and the wire carries it — which is why a theme change is a rename, not a refactor.

A design system is infrastructure, and infrastructure is judged on migration cost.

Primitive, semantic, and component tiers keep product teams one rename away from a theme change instead of one refactor away. Decisions were logged in public; deprecations shipped with codemods, so adoption did not depend on goodwill.

Downstream teams consumed the same tokens and patterns instead of re-inventing them per app.

Tiers
Primitive · semantic · component
Governance
Public decision log
Migration
Codemod-backed deprecations
Role
System author · lead
Also in the recordTwo further systems, documented
A IoT Intake FrameworkNine dependency-ordered sections that turn tribal pricing knowledge into a path any seller can run. T‑Mobile · 2025 B Continuous DiscoveryA timeboxed research practice that made 400+ inherited properties comparable, and therefore triageable. Elanco · 2022–23
Before that Earlier work, still running
  • 2021OpenELIS GlobalOpen-source laboratory information system
  • 2017Cisco Networking AcademyNetAcad 2.0; components later reused across the NetAcad site
  • 2014–16PayPalMacy’sBloomingdale’sApp and web payment integration for Macy’s and Bloomingdale’s
Evidence What the three add up to
Leadership Set the standard five designers now work to

Owned IoT Connectivity end to end, trained five senior designers on it, delegated four lanes of the work to them. One contractor’s scope became a team’s practice.

Delivery Ships under real constraint

ElancoGPT went taxonomy to production in 72 hours, on the design system and discovery process I had already built. Speed was a consequence of the system, not heroics.

Systems Owns the chassis, not just the screens

A three-tier token system with public decision logs and codemod-backed deprecations. One replicable process shared by design, business, front-end, and solution engineering: four functions, one definition of done.

Currently

Retained by product and platform teams to put AI into their products and their process, language models and vision models alike, without losing the human context it runs in. Every engagement ends in shipped software.

Open to A team shipping AI that has to hold up under real use, not just in a demo.

Twelve years on systems where being wrong is expensive: quote-to-order pipelines carrying thousands of devices per deal, laboratory records, payment integrations in live stores. I take the unclear surface, name what it has to do, and ship it with research and engineering. When the answer is not in a spec I go get it: data visualization in Python, computer vision with OpenCV, Arduino on the bench. Lead or Principal. Available now, Seattle, remote, or relocating to California.

Start a conversation I read and answer every message myself.
Projects
Pages
Actions