AI-Native Product BuilderProduct Generalist

LEE SANGINN

I connect product planning, design, engineering, growth, and experimentation to turn ideas into working products.

4 projects
Research-led projects
1 launch
Public launch
2 projects
Solo PM projects
4-9 people
Project team size

Selected work

Found the problem, then checked it myself

Public Trippixel onboarding screen for setting travel preferences

01Public launch2025.11 - 2026.05

Trippixel

AI travel coordination planner

Problem
Travelers with different preferences kept saving places, sharing opinions and fixing dates across separate tools, so the coordination fell on whoever led the plan.
Role
Solo planner/PM · 9-person team
Evidence
64 survey responses · 5 interviews · Public web launch on 2026.03.18 · GA4: 65 users / 91 session starts
Remains
The moment that earns repeat use remains to be validated.
Open live product
Moyeohaeng dashboard showing shared place collections, collaborative itinerary editing, and a travel map

02Unreleased prototype2025.08 - 2025.09

Moyeohaeng

Real-time collaborative travel dashboard

Problem
As more people joined, collecting places, gathering opinions, and settling the itinerary became increasingly fragmented across tools, piling the sorting and decisions onto the organizer.
Role
Solo PM · 8-person team
Evidence
87 survey responses · 11 interviews · User testing
Open the deck
First-round prototype screens from the Baemin one-person household proposal — order-history exposure and saved-store flows

03Prototype validation2025.07 - 2025.08

Baemin One-person Household Case Study

Independent, unofficial product case study

Problem
One-person households want to save a trusted meal and reorder it quickly, but the existing reorder and favorite flows had low visibility and did not connect saving with return visits.
Role
APM · 4-person PM team
Evidence
Interviews and usability tests · 2 prototype validation rounds
Open the deck
Dajeongi caregiver app screen — care-recipient status, step count and alarms on a single view

04Research prototype2025.06 - 2025.07

Dajeongi

Caregiver reassurance signal service

Problem
When a caregiver interprets a missed call as an accident, repeated calls leave the caregiver anxious and the older adult feeling watched.
Role
PM / Team lead · 4-person team
Evidence
6 caregiver interviews · Prototype validation with 4 participants
Open the deck

Practice

I do not stop at the plan — I build it and check whether it works.

01

I started in service planning.

I learned to find problems through research, design flows and policies, and carry a product to launch with a cross-functional team. Only after launching did I see the gap between the plan and real use.

02

I did not want to stop at the plan.

I learned design and code so I could test ideas myself. The point is not to dabble in many fields — it is to take on the roles a problem requires and carry the work through to a concrete outcome.

03

I keep the decisions behind the work.

I use AI to expand the range of research, design, building, and verification I can do — not to make the call for me. The judgment remains mine, and I do not erase the wrong calls.

Background

Seoul Cyber University
B.S. in Computer Science, in progress
GAIQ
Google Analytics Individual Qualification
goorm × Kakao Product Management Bootcamp
Completed
Jeju Halla University
B.A. in Social Welfare

Now

AI products and developer tools
I explore this by building tools where people and AI work together — the five repositories below document that work.
Turning metrics into decisions
I earned the GAIQ and connected Trippixel's metrics to real decisions. The numbers are still small, so I am still learning.
How much I can build myself
I am studying computer science again and expanding how far I can carry planned work without handing it off.

Enterprise AX workflow design

Three studies that turn work judgment into executable rules

01 · MyRealTrip · 2026.066-page caseA rule-based evaluation for cancellation-recovery exposure policyPolicy QA · experiment readiness · AX workflow design · locally verifiedShowing alternatives after cancellation can help, but repeating the cancellation reason, raising the price sharply, or lacking booking evidence can damage trust and increase support risk.02 · Musinsa · 2026.076-page caseA partner-growth bottleneck diagnosis that exposes both data boundariesTwo-sided diagnosis · data-request loop · AX workflow design · synthetic-data verifiedBrand context sits with the brand while exposure and conversion signals sit with the platform, so one side alone cannot conclusively diagnose a growth bottleneck.03 · KakaoPay · 2026.076-page caseA PR review that turns a work manual into an approvable policy packPolicy governance · evidence-based review · AX workflow design · demo-policy-pack verifiedAs AI-generated changes accelerate, consistently applying security, privacy, and operations criteria scattered across documents and expertise becomes the bottleneck.

SCOPEAll three are independent personal work with no affiliation. This work does not claim access to internal data or code, adoption in production, or measured improvement.

GitHub archive

Five tools I built

  1. 01

    Vibebuilder

    I built it to keep context, validation, and review repeatable across long-running AI development work.

    Python · Codex · Agent HarnessVibebuilder — View on GitHub
  2. 02

    iamsolazyfordesign

    I built it to turn design references into evidence-bound implementation contracts rather than surface-level imitation.

    JavaScript · TypeScript · MCP · SQLiteiamsolazyfordesign — View on GitHub
  3. 03

    Zeus

    I built it to keep AI-agent actions inside human-defined authority and approval boundaries.

    Python · Policy Gates · Local-firstZeus — View on GitHub
  4. 04

    packly-developer-preview

    I built it to route only the rules and context each AI coding tool needs for the current task.

    Shell · CLI · MCPpackly-developer-preview — View on GitHub
  5. 05

    apollo

    I built it to keep an image-to-3D asset workflow local and under Codex control.

    Python · Codex · Local-firstapollo — View on GitHub
All repositories