Eric Devlin · Selected Work

Where design systems and AI tooling meet.

Here's two pieces of work.

01 / Self-serve tooling

A generator that took the designer out of the loop

Citi · Velocity (economic research & advisory platform)

Context

Velocity's economic research analysts publish a steady stream of articles and videos, each needing an on-brand hero image. Every one of those routed through a designer first. It was a bottleneck for the analysts and low-leverage work for the design team.

What I built

A self-serve generator the analysts run themselves: enter the speakers and titles, choose from a curated set of approved backgrounds, and get a finished image that holds the brand. The constraints are baked into the tool, so the output is on-brand by construction rather than by review.

Why it matters

It removed a designer from the critical path and standardized Velocity's visual language at the same time. The same instinct sits under everything here: give non-designers a guided path to good output instead of gatekeeping it.

~5 hrs
of design time freed every week
0
designers in the loop per image
1
consistent visual standard, enforced by the tool

02 / Design systems

Reviving a firm-wide design system, built for AI to consume

Citi · firm-wide design-system unification

Context

A unification effort across the bank's design systems had been stalled for two years. The systems had drifted apart, and nobody owned the path back together.

What I did

I revived it. Co-authored the strategy, built the case, and sold it to leadership. It is now a funded platform with a twelve-person team collapsing eight systems into four.

The idea

The system is built to be consumed by AI tools, not just read by humans. We encode brand and interaction rules (voice, density, behavior per line of business) so an AI can generate the right interface for the right context. I think that is where design systems are going, and I want to build it somewhere it is the product rather than an internal initiative.

8 → 4
design systems being unified
12
people on the funded platform team
2 yrs
stalled, then revived and funded