I design trust and AI systems for regulated marketplaces, where value passes from business to business to the people who actually rely on it.
Selected work
Established trust infrastructure across a marketplace of 400K+ agents and 150K+ loan officers.
When a platform acts on your behalf, failure isn't a UX problem, it's liability. We were managing existing referral relationships and initiating cold outreach, both under strict RESPA constraints. I turned that compliance burden into a user-facing system: distinct experiences for agents and loan officers, with the system enforcing consistency at scale.
Replacing email open rates with a signal LOs could actually act on.
An approved analytics dashboard told LOs their emails were getting opened. It didn't tell them who to call. I replaced open rates with Lead Status — a signal LOs could actually act on — and named leads converted at more than double the rate of a generic summary link.
How defining what AI was opened up new ways to design it.
When Company C introduced Mila, an AI assistant and mascot, "AI" became an umbrella term and it got harder to tell what was actually AI. I built an attribution model separating deterministic systems from AI, then used that clarity to define how AI should look, behave, and disclose itself differently from the rest of the product.
Reduced AI system cost by 95% and made it viable for production use.
Company C's design AI wasn't scaling. At roughly 180,000 tokens per scan, a few runs could exhaust a weekly budget. I introduced a deterministic pre-analysis layer to control what reached the LLM, then restructured the system's decision-making with routing, orchestration, and evaluation layers to make outputs predictable and controllable. Once that foundation was in place, I added a judgment layer that taught the AI how to evaluate designs, not just identify them.
More work
Case study coming soon
After an acquisition, I worked across two platforms and their suites of tools to create a more coherent experience. That meant sorting through overlapping functionality, deciding what to keep or consolidate, and figuring out how the combined platform should work going forward.