Most AI products ship with interfaces only the data science team can operate. I design the other kind. For the past year at Aampe, I've been shaping the product surface of an agentic messaging platform, where autonomous agents decide what to send, when, and to whom, and learn from every response.
My work sits between the people who build these systems and the people who use them. I turn model behavior into controls a non-technical user can trust.
Aampe's AI agents optimized millions of messages but measured success the same way for every customer. I designed the system that lets marketers define what "good" means for their agents, turning a black-box ML signal into a self-serve configuration flow.
See Overview
An AI messaging platform's agents optimized millions of sends per week but had no concept of business priorities. I designed the control that let marketers steer agent behavior without overriding it: shipped V1, audited its failures across seven dimensions, then redesigned V2 around one insight that changed everything.
See OverviewA mobile app designed to optimize loan recovery operations. Enabled 100+ field agents to efficiently manage loan collections and incentives, improving recovery rates and transparency across teams.
Read NowTransformed the manufacturing process by solving inefficiencies tied to paper-based instructions, resulting in faster information flow, reduced manual errors, and measurable gains in productivity.
Read NowVera is an iOS app that distills 11 Apple Watch signals into a single biological age, revealed once a week like a ritual. I designed and built it by assembling a team of AI agents through Claude and Codex. No engineering team. One designer running an AI studio.
Read Now