TrueML: a design system built for AI-era prototyping
TrueML builds digital debt collection products, which means every screen sits in a highly regulated, emotionally loaded context. Three product teams were shipping consumer, client, and internal interfaces in parallel. My job wasn't only the design system - I designed many of these products end to end myself, while also giving the teams one shared foundation that evolved alongside the work that mattered most. The system was never "done": each redesign tested it, and each improvement to it sped up the next redesign.
The design system
An enterprise design system in Figma - variables, tokens, and a full component library - kept in sync with a shared React library through Code Connect. It covers everything from atomic elements to complex UI, and it's shared across three product and dev teams.
- Challenge
- Three teams, three products, and a design-to-code handoff that drifted every sprint.
- Approach
- One source of truth in Figma, structured to match the React library one-to-one, with Code Connect linking the two.
- Key decision
- Governed but flexible. We built one token and component system that every designer and developer could use and extend within clear rules.
- Outcome
- Faster, cleaner handoffs across three teams. The system became the base for all the product work below.
A prototyping kit for AI tools
As the system matured, I built an AI design framework and a brand-specific component kit that works in Figma Make, Claude Design, and other prompt-to-UI tools. Product teams could generate on-brand, system-accurate prototypes themselves instead of waiting in the design queue.
- Challenge
- Demand for prototypes outgrew the design team, and headcount wasn't growing.
- Approach
- Package the system so AI tools could use it correctly, and give PMs and engineers a framework for prompting it.
- Key decision
- Don't bet on a single tool. I translated tokens, components, and guidelines into markdown files and code, so the system works with any current or upcoming AI design tool.
- Outcome
- Dramatically increased prototyping capacity across product teams with flat headcount.
Consumer payment flow A/B test
The consumer-facing flow where people set up payment plans for their debt. We tested UX changes in production, kept what worked, and moved the flow fully onto the design system.
- Challenge
- A high-stakes flow for an anxious audience, plus a growing maintenance cost from the legacy UI.
- Approach
- Customer interviews and usability studies with PMs and researchers, then iterative testing of individual UX changes.
- Key decision
- Customers in financial stress prefer clean UI and clear up-front messaging when presented with options. Fancy designs aren't helping.
- Outcome
- +5% conversion to payment plans, and lower ongoing maintenance cost after integrating with the design system.
Client onboarding
Onboarding new clients used to cross three separate interfaces: the consumer experience, client tools, and internal operations. Much of it lived in backend tools with dated UI that only a few people knew how to use. I designed one experience that connects all three.
- Challenge
- Onboarding took too long because it was split across tools and teams, and every client needed hands-on help from specific internal team members.
- Approach
- Map the whole journey across consumer, client, and internal interfaces, then design them as one flow.
- Key decision
- Replace specialist hand-holding with guided self-serve. Specialists also moved into the same interface clients use - just a different admin level, not a separate internal tool.
- Outcome
- 3x reduction in client onboarding time.
Internal and client tools
The less glamorous backbone: login and authentication, user management, and account tools for internal admins. Dense, permission-heavy interfaces that people use all day.
Multichannel messaging CMS with AI assist
A tool for clients to write the messages they send to consumers by email, voicemail, and SMS. One WYSIWYG editor covers all three channels, with pre-built and client-created variables, plus AI-assisted content creation.