Limin Zheng
Product manager. Former product designer. Independent builder.My background spans complex B2B products, cloud security, and regulated systems. More recently, I’ve been building AI products end to end.
- Status
- Live
- Form
- Web app
- Related
- Roleward Job Hunting — agent skill
An AI job-search workbench: finding roles, judging fit, preparing applications and tracking progress, all grounded in your real work history. Facts, inferences and drafts are always kept apart.
- Status
- Live
- Form
- Web app
Sorts, clusters and summarizes a batch of user feedback into traceable work packs a team can pick up. Also a complete, end-to-end exercise in eval design.
- Status
- Live
- Form
- Web app
- Related
- Score Simplifier — agent skill
A practice companion for adult piano learners, to make limited practice time count. My first product: v1 was a minimal feature set built in Lovable; phase two moved to Claude Code.
- Status
- Live
- Form
- Web app
Twenty minutes to leave your family an index of what they would need. A privacy-first handover tool — for the day something arrives before “tomorrow” does.
- Status
- Alpha · v0.5.11 · MIT
- Runs in
- Codex, Claude Code, Claude web
Turns your conversations with AI into material for reflection: diaries, weekly reviews, assumptions questioned, prompts for self-exploration, and learning worth keeping. Works in Chinese and English.
- Status
- Alpha · v0.1.0-alpha.3
- Runs in
- Codex, as a local workspace
- Related
- Roleward — product
The decision method behind Roleward, run as a local Codex workspace — fewer, better-fit applications instead of more of them.
- Status
- Alpha · v0.3.8 · MIT
- Runs in
- Codex, Claude Code, Claude web, ChatGPT Work
- Output
- PDF, MusicXML, MIDI
- Related
- Pianology — product
An AI skill that makes an easier solo-piano arrangement from a reliable source score, at three levels, and keeps refining it as you practise. The aim is not fewer notes but less real playing burden — reach, jumps, hand conflicts — while keeping the piece’s musical identity.
- Where
- Substack — essays, notes, and the Indie Notes series
- Latest
- Sep 22
The long-form side: arguments worked through properly, usually starting from something in the daily notes.
- What Companies Actually Mean When They Hire an “AI Product Manager”A qualitative review of 43 current roles, and a map of the very different work now hiding behind one title.
- AI Can Save Time Without Saving MoneyWhat recent enterprise evidence tells us about cost, value, and the gap between productivity and ROI.
- Season 1
- 100 / 100 — complete
- Season 2
- 039 / 100 — since Sep 5
- Languages
- Chinese, English
One note a day across seven topics: understanding AI, product and design, building solo, working with AI, human and AI, career, and life. First posted on RedNote, archived in Chinese and English.
Hi, I’m Limin.
I spent years as a product designer, then a product manager, on complex B2B products, cloud security, and regulated systems — work where a wrong answer costs something.
Now I build AI products on my own, end to end. The thread through all of them is judgment: what AI should decide, what it should only suggest, and how to tell the difference.
Things I’ve made
No. 1LiveRoleward
A workbench for finding better-fit roles.
Facts, inferences and drafts are always kept apart.
An AI job-search workbench: finding roles, judging fit, preparing applications and tracking progress, all grounded in your real work history. Facts, inferences and drafts are always kept apart.
No. 2LiveAsterline
Turns user feedback into traceable work packs.
An end-to-end exercise in eval design: defining what “right” means.
Sorts, clusters and summarizes a batch of user feedback into traceable work packs a team can pick up. Also a complete, end-to-end exercise in eval design.
No. 3LivePianology
A practice companion for adult piano learners.
My first product — built with Lovable, then Claude Code.
A practice companion for adult piano learners, to make limited practice time count. My first product: v1 was a minimal feature set built in Lovable; phase two moved to Claude Code.
No. 4LiveAllGood
A privacy-first index of what your family would need.
Twenty minutes, one index for your family.
Twenty minutes to leave your family an index of what they would need. A privacy-first handover tool — for the day something arrives before “tomorrow” does.
Reflection Companion
Diaries, weekly reviews, questioned assumptions — a skill for Codex and Claude.
GitHub ↗Roleward Job Hunting
Roleward’s method as a local Codex workspace.
GitHub ↗Score Simplifier
Start with the music you love — an easier solo-piano arrangement that keeps the piece’s character.
GitHub ↗What I keep thinking about
What should AI decide, what should it only suggest — and how do you tell?
- 01What Companies Actually Mean When They Hire an “AI Product Manager”A qualitative review of 43 current roles, and a map of the very different work now hiding behind one title.Notes from the In-Between · Aug 30
- 02AI Can Save Time Without Saving MoneyWhat recent enterprise evidence tells us about cost, value, and the gap between productivity and ROI.Notes from the In-Between · Sep 11
The 100-day habit
A short note every day, on RedNote and 100 Days Building. Every box is a day; the colour is what it was about.
What a cup of coffee does to your brain — life & reflections
