HeyMimoo is a pet memory brand I built from 0 to 1 as a one-person company, with AI involved across branding, technical execution, and GTM. In its first four months, it generated 307K+ views and 23 paying orders.
(1) Model choice
Tested 3 generation approaches; frame-by-frame looked choppy in realistic styles, so I chose Dreamina for smooth, lifelike motion.
(2) Pricing
AI costs left early orders with almost no margin, so I split the offer into a standard and a custom tier.
(3) Model drift
When updates broke reliable prompts, I simplified them and rebuilt the workflow around what still worked.
Fleeton is a T-Mobile sponsored system that reroutes last-mile drivers when traffic or breakdowns put deliveries at risk. I shaped the interaction design and owned the entire frontend, all the way through backend integration.
(1) Alert Rules
Defined three alert tiers by impact on delivery windows: stay silent, reroute, or escalate to dispatch, so drivers are interrupted only when it matters.
(2) Debugging
When old alerts kept resurfacing, AI fixes went in circles. I read the logs, traced it to the event stream replaying from the start, and fixed it with cursor tracking.
(3) Where AI Didnโt Belong
We tested AI against a traditional routing algorithm. For routes with strict delivery windows, the algorithm was more reliable, so we kept AI out of rerouting.
A take-home challenge from my PM interview with Reframe, an alcohol-reduction app. The brief: reimagine the onboarding from scratch to reduce drop-off, including the flow, branding, and visuals.
(1) Brainstorm Together
I started with my own analysis, then Claude offered a wide range of ideas fast. I picked the ones that clicked and dug deeper, turning brainstorming into a quick back-and-forth.
(2) Design Through Conversation
I directed every screen while Claude built them as we talked. With execution this fast, I could test different directions in minutes.