A researcher of people and stories in the age of AI
AI is not just a productivity tool.
I use it to build infrastructure for self-knowledge, emotional order, and the inner life.
I build products, write, and make a podcast to explore how AI can help people act, express themselves, and understand one another more clearly. I am drawn to things that feel vague, invisible, or hard to name, and to giving them evidence, language, and a structure that helps someone move forward.
Based in Shanghai.
Turning uncertainty into a next step.
Job searching is the most mature application so far, but the underlying question is broader. Can AI help people weigh evidence, recognize constraints, and turn a complex reality into a concrete next step?
A private system that narrowed 1,667 public job records to 10 actionable roles. It checks location, experience, and seniority gates first, then connects personal evidence, role-specific materials, and a next step into one traceable action chain.
A decision workspace for small hospitality renovations. It connects source materials, constraints, spatial diagnosis, option routes, risks, and decision history into a traceable loop. The public version uses independent sanitized data and clearly labeled deterministic rules.
It detects evidence gaps across numbers, ownership, causality, AI mechanisms, scale, and execution, then uses dynamic follow-ups to build evidence cards, faithful rewrites, and an interview preparation pack. The public version contains synthetic data only; real-user validation is still pending.
Turning human taste into standards a model can learn from.
I care about judgments that are difficult to quantify. Why does one sentence feel alive, one response make someone feel understood, or one essay still sound like its author? I am exploring how those human judgments can become standards for evaluation, learning, and continuous improvement.
Starting with blind language review, it breaks down why one response works better across intent alignment, emotional attunement, narrative pacing, user agency, and safety boundaries. It is currently a strategy prototype, not a finished work.
A Newsletter thought studio designed to protect reading continuity and the author's voice. Topic reports, translated full text, highlights, deep discussion, a thought canvas, and spoken-draft output form one continuous path. It is not yet connected to real Gmail or the Claude API.
Helping people feel seen while preserving boundaries.
This direction begins with feeling, relationships, and the experience of being seen. I want to know whether AI can help people understand themselves and one another more accurately, and respond with warmth without overstepping. That experience may live in a conversation, a voice, a color, or a visual world.
Enter two birthdays to create a relationship palette and a shareable card. The focus moves from compatibility to a gentler question: what does this attraction reveal about me? Color becomes a lightweight mirror for relationship experience.
A phone call between Shanghai and Guangzhou about work, relationships, and the quarter-life passage. Published episodes move from a real layoff story to the human capabilities that remain valuable in the age of AI, while practicing what it means to hold one another through honest conversation.
An Obsidian vault holds the source of truth for thinking, Notion handles daily dispatch, Claude Code and Codex support research and building, and every product has its own workshop. The tools may change. These rules do not.
If you are building human-centered AI products, content, research, or evaluation systems, I would be glad to hear from you. I read job opportunities, collaborations, and thoughtful conversations carefully, and I usually reply.