Jiayi Liu

Acorn Aware (Bestie AI)

• Multi-agent generation system: with the growth and engineering teams, moved ad creative from manual Figma work to multi-agent generation with human review. I owned the flow framework: three image agents for conservative variants, new concepts and competitor remakes, and how they hand off; the variant agent generates by "5 hypothesis dimensions × 1–3 options", producing A/B sets that can genuinely be compared. A full asset set for one concept across four markets went from about 3 days to about 1.5 days.
• Writing the brand spec as rules for AI: wrote the image agents' prompts; diagnosed generation drift such as avatars misused as logos and mixed brand elements, and rewrote the visual spec as generation constraints and review standards AI can enforce, shipped in pairing with engineering, cutting manual rework after generation.
• Human-in-the-loop: helped design four internal tools (asset library, prompt editor, review queue, competitor browser), deciding at which step a person steps into the agent flow, what they look at and what they change; prototyped them in the company's real codebase with AI-assisted development before engineering built them out.
• AI relationship understanding system (0 to 1, live): rebuilt a CRM-style contact list into a system where AI keeps track of the people in a user's life, how relationships change and where the emotional risks are. Defined from scratch the person data model AI maintains, then turned AI's understanding into a relationship orbit and radar you can read at a glance, showing which relationships are cooling.
• Designing payment around AI cost: to control AI chat costs, worked with the PM to move from weekly/monthly/yearly plans to four usage-based membership tiers; owned the paywall, limit prompts and upgrade paths, trading off cost against experience, and iterated through several versions on data after launch.
• Website: designed and built the Bestie AI website on my own with Cursor / v0, from kickoff to launch in 3 days · bestieai.app

Sabbatical Lab

• Designed the core AI content-creation flow (generate → edit → export), defining the template framework and visual spec for generated results, and built the foundational design system (color tokens, typography, component states).
• Designed social media and ad content generation templates, strengthening the platform's visual tone and brand consistency.

Cancer Hospital, Chinese Academy of Medical Sciences

• Led user research and workflow redesign of the hospital's discharge follow-up system; mapped core pain points from interviews with 18 clinicians and 25 patients, then iterated through three rounds of testing with real users.
• Designed accessible UI components for elderly users, iterating over 3 rounds on button hierarchy, text legibility and step breakdown.

• Experience design for the Luojiang overseas-Chinese digital living-room platform; contributed the user-experience section of the Future AI Aesthetic Education white paper.

• Experience and interaction design for the official VR exhibition platform of the 20th National Congress; led visual delivery of the website and VR scenes, and designed a virtual character system based on the costume cultures of 52 ethnic groups.

An emotional-support wearable for cancer patients. Owned the interaction experience and companion UI, using emotional feedback to ease anxiety and loneliness during treatment. Berkeley Outstanding Graduate Design Award.