AI naming plus tournament-style voting — so the whole family picks the baby's name together

When the time came to name a baby, I tried out a few naming services. Some used AI to suggest names, others applied traditional Korean naming rules — but after using them all, one shortcoming kept coming up. They were all "single-player tools".

In real life, naming a baby isn't something you do alone. Mom and Dad weigh the candidates, show them to the grandparents, who say "hmm, not quite...", and then aunts and uncles chime in too. The name only settles once the whole family's opinions come together, but existing services ended at "one person gets an AI suggestion, screenshots it, and drops it in a KakaoTalk group chat."

So I thought — what if we keep the AI-generated names, but layer on a tournament-style voting (bracket voting) format? Compare candidate names head-to-head in 1v1 matchups, let each family member vote, then tally the results to land on the final name. AI recommends, family chooses.

• Existing naming services are all single-player

  Whether AI-driven or tradition-based, every tool delivers results to one person at a time. None of them had a way to gather opinions from the rest of the family.

• Naming is supposed to be a family event

  In Korean culture, not just parents but grandparents and relatives weigh in — and no service was built to accommodate that.

• The only way to share candidates was a KakaoTalk screenshot

  To show AI-suggested names to the family, taking a screenshot and dropping it into the group chat was the only option.

The structure is simple. Create a project, enter the baby's details, and GPT-4o suggests names. Open a bracket-style tournament with those candidates, share the link with the family, and each person votes. As the votes roll in, the family's favorite name naturally rises to the top.

1. 01

   GPT-4o prompt design — teaching the AI the rules of Korean naming

   Korean traditional naming has a surprising number of rules. The generative and destructive cycles of the Five Elements (木火土金水 — Wood, Fire, Earth, Metal, Water), the Saju Palja birth-chart calculated from the baby's birth date and time, and the real-world usage frequency of each Hanja character that maps to a given syllable. I baked all of that into the system prompt so GPT-4o would produce names that feel like they came from an actual Korean naming studio.

   One thing I spent extra effort on was filtering out the Hanja that GPT naturally gravitates toward (like 瑞 or 睿). Models reach for these characters often, but they're rarely used in real Korean naming. I explicitly banned those, and locked the output format to JSON to keep parsing stable. The prompt itself is core service logic so I'm not publishing it here, but the overall direction is "structure Korean naming rules into a rule list, state forbidden patterns explicitly, enforce a JSON schema."

2. 02

   Bracket engine — a client-side tournament

   A tournament bracket that supports 4, 8, 16, and 32 entrants runs on the client. The candidate names get randomly shuffled into a bracket, and each round is a 1v1 comparison where picking one advances it to the next round.

   Once the share link goes out via KakaoTalk, family and friends can jump straight into voting. Each person's votes are stored in Supabase, and the project owner can see the aggregated results. The free plan supports up to 10 voters per project.

3. 03

   Hanja meanings + Ohaeng data

   I mapped Hanja variants for 40+ Korean surnames. For example, even the surname "Kim" can be written with different Hanja like 金 or 金海, and each variant has a different stroke count and Ohaeng (Five-Element) attribute. Without that dataset, there's no way to give GPT-4o concrete instructions like "this surname's element is Metal, so pick generatively compatible Water or Wood Hanja."

   On the results screen, each syllable's Ohaeng is color-coded. Wood is green, Fire is red, Earth is yellow, and so on — so the elemental composition of the name is readable at a glance.

4. 04

   Kakao login + sharing — designed for virality

   Kakao OAuth is wired into Supabase Auth. Since most Korean users already have a Kakao account, the login friction is minimal, and sharing the voting link as a KakaoTalk message turns the product naturally viral.

   When voting ends, a postcard-style result card is generated, complete with an 印章 (seal) design. That card is also shareable via KakaoTalk, so families can use it to announce, "Here's the name we picked!"

• ohmybaby.kr is live

  A side project I took solo from planning through AI integration, UX design, and deployment. It is running as a real service today.

• Translating Korean naming culture into AI prompts

  The biggest takeaway was turning domain knowledge — Ohaeng generative/destructive relationships, Saju analysis, and filtering of Hanja actually used in real naming — into prompt engineering. I learned that "what you forbid the AI from doing" matters more than "what you tell it to do."

• End-to-end experience: planning, design, build, deploy

  Stacking Next.js 16 + Supabase + GPT-4o + Kakao OAuth full-stack on my own gave me experience with auth, billing design, AI integration, and social sharing inside a single project.

• Next up: activate the Pro plan, expand marketing channels

  The Free/Pro tier structure — gated on AI suggestion count (free: 3) and vote-participant count (free: 10) — is already designed. The plan is to collect user feedback, turn on the Pro tier, and broaden marketing channels through parenting communities.