Five self-built AI products, and one creator-ecosystem case study with real data.
AI Product Strategy ／ Creative Tools & Creator Ecosystems ／ MSc, Creative Industries (UK)
I do not believe you can build a good product without touching real users. Over the past year I shipped five AI products on my own, and in a 91,000-member overseas community I turned creative participation into something the data could verify. Everything below can be opened and used, and every number traces back to a source, including the ones that do not flatter me.
Ongoing writing: 爱吃温泉蛋 (emotions & self-observation)·紫海盐 (AI & ordinary lives)
No screenshots. Open them and use them.
Five products dealing with printing, relationships, decisions, evidence and space. They answer one question: when judgment resists quantification, can AI still give an answer that is traceable, bounded, and open to being wrong? Each states its real maturity, including what it does not yet do.
What it does: restores the physical defects of risograph printing to digital images: colour separation, halftone dots, misregistration, paper grain. Everything runs locally in the browser; no image is ever uploaded. The hero image above and the two plates below all came out of it.
What it does: generates a relationship palette and a saveable card from two birthdays, using colour as a light mirror for relational experience.
What it does: breaks "does this claim hold up" into six gaps (numbers, ownership, causality, AI mechanism, scale and delivery), then uses adaptive follow-ups to build evidence cards and an interview prep pack. Claims without support are downgraded or refused. 5 real users have trialled it with 20+ redacted and synthetic cases; systematic validation is not complete, and the public build uses synthetic data only.
Narrows 1,667 public job records down to 10 currently actionable roles. Hard gates like location, years of experience and seniority are checked first, then personal evidence, role-specific materials and a concrete next step are connected into one traceable action chain. 204 automated checks passing.
Connects documents, constraints, spatial diagnosis, design routes, risks and decision records for small hospitality renovations into one traceable loop. The public version uses an independent redacted example, and the current analysis runs on clearly labelled deterministic rules.
The first starts from double-blind language evaluation, breaking "why is this sentence better" into intent fit, emotional uptake, narrative pacing, user agency and safety boundaries. The second is a newsletter studio that protects reading continuity and the author's voice. Both are still prototypes and are not presented as finished work.
In a community of 91,000, only 117 people actually spoke.
I ran the overseas community and creator ecosystem at an AI agent company. This case is not a list of campaigns. It records how I used mechanism design to turn the hundred-odd people who really existed inside an inflated community into content supply the product needed, and how the data, once the mechanism stopped, proved what it had been doing.
Every claim maps to an internal retrospective, platform analytics exports, or archived channel records. Unreleased commercial and product-roadmap details are deliberately excluded.
A system should catch you, not discipline you.
I am not a highly disciplined person. Low energy, scattered thinking, and a perfectionist streak. Over ten years I tried every mainstream knowledge-management method and none of them ever really ran, because they all assume someone willing to maintain them daily, and I am exactly the kind of person that assumption filters out. So this system starts from one premise: hand all the maintenance cost to AI, and keep only input and judgement for myself.
In one line: give the world to the machine, and keep yourself.
Full method: where the four rules came from, the traps, and what is still open →
Before I cared about AI, I cared about how people express themselves and get understood.
My training is in literature and the creative industries, and I have worked across film IP, art fairs and communications. That shapes how I read users: I look at motivation, emotion and circumstance, not only conversion rates in a funnel. The two things below are ongoing practices in expression, and the upstream source of every product idea.
A friend and I started with a phone call between two cities, talking about work, relationships and the quarter-life crisis. Published episodes range from being laid off to the skills that still matter in the age of AI, and are also practice in catching each other through honest conversation. I lead positioning, topics, outlines, recording, editing and promotion.
紫海盐 writes about AI and ordinary lives; 爱吃温泉蛋 writes about emotion and self-observation. Representative pieces discuss the intuitive builder, and where a person's history and worth sit in the age of AI. This is where observation becomes language, and the upstream source of every product idea.
Rather than "open to opportunities", here is what I am actually looking for.
If you work on any of these, or simply want a serious conversation about AI and people, please write. I read job opportunities, collaboration invitations and genuine discussion carefully, and I usually reply.
juec9316@gmail.comFull CV (EN／ZH) on request, usually same day
