A founder-built AI creation product
I led Memora from user interviews and 205 product screens to a working agentic video prototype, beta users, and a live San Francisco launch.
THE PRODUCT IDEA
Click through the product flow, or replay one preserved pipeline run made by the real 2024–25 prototype.
↳ Click a stage to open it
A user selects the photos and clips that actually belong to a memory—not an endless editing timeline.
Input · photos, clips, contextSAVED PIPELINE RUN · CAVEAU BOSTON
This recruiter-safe demo uses the original inputs, selected scenes, narration plan, and 37-second render. It replays the decisions the system made instead of pretending to generate a new video.
11 source assets · 8 selected scenes · 00:37 final cut
205 SCREENS, ONE SYSTEM
How little should the user have to edit?
The answer became a simple loop: capture, create, review, share, and revisit. I mapped the information architecture, core journeys, social interactions, and AI handoff in Figma.
Browse the complete design fileGENERATED OUTPUTS
Real footage in. A captioned, narrated memory out. These were early experiments in pacing, story selection, and the tone of an AI-assisted editor.
Prototype walkthrough
A working snapshot of the original product: choose media, generate, and watch the resulting Memora. The production APIs are archived; this video preserves the real interaction.
BUILDSPACE · SAN FRANCISCO
I flew to San Francisco and presented Memora directly to 1,000+ founders at Buildspace. The product moved from a private build to something strangers could see, question, and try.
SF / 2024
WHAT I PERSONALLY OWNED
All UI/UX, product flows, agent roles, and the creation ecosystem.
User interviews, beta recruitment, feedback loops, and product decisions.
Launch videos, demos, posters, brand direction, filming, and editing.
A five-person cross-functional team spanning the US and China.
Pitch decks, investment conversations, events, and the SF presentation.
WHY THIS MATTERS FOR AI CREATION PRODUCTS
Creative AI succeeds when model capability becomes a clear interaction, a measurable quality bar, and a reason to return. That was the product problem I worked on for a year.
Turned an agent workflow into a five-step creation experience.
Used beta feedback, satisfaction, and editing-time reduction as product signals.
Translated between users, designers, frontend, backend, and ML thinking.
Built with a US–China team and worked bilingually in English and Mandarin.
SwiftUI frontend · Python backend · LangGraph agent orchestration · MoviePy/FFmpeg rendering. The original system is archived; the product evidence on this page is preserved from the real build.
I’m still obsessed with one question: