CURRENTLY IN PROGRESS
Benchmarking vision-language models to catch falls in video — built for a real client engagement
Python, GPT-4o, Claude, Gemini, Qwen2.5-VL
A desktop app that analyzes video for fall events using vision-language models, built with a small team over a 3-month client engagement. Rather than betting on one model, it benchmarks GPT-4o, Claude, Gemini, and Qwen2.5-VL side by side so the client could actually see the trade-offs before committing to one.
DESIGN PROCESS
Onboarding is split into single-question screens with a progress bar, based on the idea that people commit more easily to a short, predictable path than an open-ended form.
DESIGN PROCESS
Condition badges and verified-seller marks appear directly in the browse feed, not buried in item pages, since trust is the main thing standing between a browser and a buyer in secondhand shopping.
DESIGN PROCESS
Buy Now and Add to Cart are separated deliberately. Most items here are one-of-one, so buyers who already know they want it shouldn’t be forced through a cart step.
CONCEPT — NOT YET BUILT
Early exploration of a full checkout flow using familiar payment patterns (Apple Pay, Google Pay) to reduce hesitation at the final step. This is a design direction, not a shipped feature yet.
01 / IN THE BUILD
Three seller roles
Individual closets, vintage stores, and independent creators each have a clear path in.
Browse to purchase
The flow covers browsing, item pages, cart, checkout, and public seller profiles.
Trust up front
Condition, seller verification, and reviews arrive early—where secondhand decisions happen.
02 / NEXT MILESTONE
The next layer pairs express payment with a creator mode, then puts Minty in front of a real vintage store to learn from actual transactions.