JEONG (SOLHUN) — Gyeonghun Jeong

Every screen fans out into the app, mobile web, desktop web, teacher web and admin web. Matching screens with the designer and the client was a bigger job than the code itself, so together with the designer I rebuilt the feedback process that kept losing items when requests were vague or piled up.

1. Responsive review

   Before

   Dragging the Chrome window wider and narrower, one screen at a time

   After

   A review board showing 55 screens and 39 modals in portrait and landscape at once. Each screen has a comment box, so a designer’s comment becomes a work instruction as is.

   55 / 55 tablet screens checked · 0 errors

2. Sending feedback

   Before

   Sixty text notes in a single Slack message

   After

   A widget that lets you point at the actual button or text on screen and report it, added to all four apps: app, web, teacher and admin.

   41 reports the designer filed in a single day

3. Tracking progress

   Before

   A 613-line review doc updated by hand; you had to ask to know what was fixed

   After

   Report → internal system → Linear issue → Slack notice, all automatic. When a fix ships, Slack gets what changed plus a link to the review board.

   301+ reports with a tracked status

I worked with the labor attorneys as if we were one team. We moved checklist-and-document manual work into software and shifted the consulting process to AI (AX). It also checks the statutes an AI cites against the actual law.

1. Everything around the code

   Before

   Outsourced development usually covers the build only

   After

   Contract (Mar 8) → feature spec v2–v2.2 (Apr 13–17) → admin and advisor guides (May 17–26) → design system handoff (Jun 8) → proposal, product video and card news (Jul–Sep), all done by me.

2. Screens for each role

   Before

   Everyone saw the same data on the same screen

   After

   CEO, executives, advisors, case handlers and staff each see different data and screens. Designing that and walking the attorneys through it taught me a great deal about HR and labor work.

It started with my own mess: ten terminals open and no idea which was which. Ten months and 56 releases later it’s a public product. I did the planning, development, releases, landing page and marketing alone — it launched as a paid app, earned its first revenue, then went open source.

1. v1.9 · AI Control API

   Before

   With non-interactive runs, you only find out an AI went off track once it’s done

   After

   A local API lets another AI open sessions and work inside CLI Manager. AI sessions show up in green, so you can check in and step in at any time.

I built it because people were babysitting recurring issues by hand — checking progress, reviewing, asking again. Symphony picks up an issue, has AI agents build it from the docs, and pings a human on Slack when it’s stuck. Docs, test cases and reports all flow through Lassorun.

An admin system handling 17 quiz types × 16 question types on live production data. Questions were locked inside quizzes and couldn’t be reused, so I redesigned it so questions live on their own and link to quizzes — then verified the production data migration step by step, from design and plan to execution.

1. QA

   Before

   Clicking through screens one by one to find bugs

   After

   A table that round-trips every type through create → read → update → delete and checks the saved values come back unchanged. I also tried an AI QA tool first and shared it; two teammates applied it to their own areas right away.

   17×16 CRUD round-trip matrix, 5 h 41 min

The project where I handled the most post-launch operations — payments, refunds, store review and security.

1. Payments

   Before

   Settlement broke every time payment gateway rules changed

   After

   Switched gateways four times, redesigned settlement, and landed on in-app purchases.

2. Refunds

   Before

   Users spent credits and then got refunds (refund events weren’t received)

   After

   Within 3 days: a refund webhook plus automatic credit clawback.

3. Release

   Before

   Two App Store rejections, and every fix meant another review

   After

   Moved to OTA updates so fixes no longer wait on review every time.

4. Security

   Before

   Security gaps in code built quickly with AI agents

   After

   A full audit four months in blocked 4 critical issues.

I led the redesign and ran the loop for designer and client feedback. I set things up so the designer could install the feedback tool on their own, batched incoming reports, and shared what shipped as a list — “6 reports from yesterday and today, fixed”. It carried over the feedback process refined on KITS.

1. Applying feedback

   Before

   Change requests scattered across messengers and documents

   After

   Reports left right on the screen → fixed in batches → shared as a list of what changed

Collect popular posts on X → two AI agents debate the topic (up to 3 rounds) → research, fact-check, copy, render and caption → publish to Threads, X and Instagram at once. Every day it gathered engagement data and remade the best posts from a new angle. I open-sourced it with the API keys left blank.