Taewon Kwon

Gomi

A map of places someone you trust vouched for. A Flutter app for iOS, shipped to a closed beta, and the documentation system that kept an AI-assisted build coherent for months.

Role
Founder: product, design, engineering
Type
In-house consumer product
When
2026 [TK: start month] – Aug 2026
Status
Discontinued Aug 2026
Stack
Flutter, Dart, Kakao Maps, iOS
Beta
15 testers on TestFlight, summer 2026

Star averages are noisy and know nothing about your taste. The recommendations people actually trust are buried in group chats. Gomi made the unit of content a place someone you trust vouched for, and built discovery on that social graph, on a map.

The problem

Built for the Korean market. When you look for somewhere to eat, the public signal is an average of strangers’ ratings. The signal you’d actually act on, a friend saying “go here, get this,” is scattered across chats and hard to find again when you’re standing on a street deciding.

Train of thought

The filled markers are where the plan changed.

  1. The ideaReplace the star average with the vouch: a specific place, endorsed by a specific person you follow, with a photo and a note.
  2. Map firstFood discovery happens somewhere. Putting vouches on a map with Kakao Maps answers “where near me,” not just “what’s good.”
  3. Building alone, with AIA solo, AI-assisted build moves fast and drifts fast. Before more features, I built a documentation system so every session started from the same written truth.
  4. Beta, summer 2026Shipped to 15 TestFlight testers. The two hardest problems were structural: the double cold start, and the same restaurant entered several different ways.
  5. Aug 2026Discontinued. [TK: the reason, in one or two sentences. What did the beta show, and what would have had to be true to continue?]

What I built

  • A Flutter app for iOS with Kakao Maps: map-first discovery plus social food logging.
  • A data model of users, places deduplicated against the map provider, vouches with photos and notes, saved collections, and follows.
  • Shipped as a closed TestFlight beta to 15 testers in summer 2026.
The vouch is the unit. Everything on the map traces back to a person you follow.
Image: three app screens side by sideMap, a place with its vouches, and a profile [TK]
The app. [TK: caption]

Hardest problems

  1. Double cold start.Gomi is a social product and a content product at once, so a new user with no follows and no nearby posts sees an empty app. [TK: what you tried for it.]
  2. Place canonicalization.The same restaurant entered several ways splits its vouches and breaks the feed. Places are deduplicated against the map provider. [TK: the cases that still slipped through.]

How I built it

The part of Gomi I’d bring to any team. AI-assisted development is fast, but every new session starts with no memory. The fix was to make the project’s state written, canonical and read back before any work.

  1. Session starterForces a read-back of current state before any work
  2. BuildAI-assisted coding, with diff review on every change
  3. CloseoutWrites targeted updates back into the docs
  4. Canonical docsProject state, deferred issues, design tokens
The session loop. Every session starts from the docs and ends by updating them.
  • One project-state file as the single source of truth.
  • A deferred-issues log where every parked item has an explicit revisit trigger.
  • Design tokens for consistency without a designer.
  • AI for architecture and planning, and AI-assisted coding with diff review on every change.
Image: an excerpt of the project-state fileOr the deferred-issues log, with a revisit trigger visible [TK]
The source of truth. [TK: caption]

Outcome

Shipped to a closed TestFlight beta of 15 testers in summer 2026, and discontinued in August 2026. [TK: why, and what you learned from the beta.]

What it shows: shipping a consumer mobile app end to end, and running a disciplined AI-assisted development process that stays coherent over months.