Skip to content
Back to work
AI consumer fortune-telling SaaSLive

간지사주

A 256-page mobile fortune-telling platform with Toss payments, a coin economy, and idempotent refund logic — grounded in a classical-text RAG corpus with 6 daily push dispatches.

간지사주 — AI consumer fortune-telling SaaS

Overview

Fortune-telling is a notoriously soft product category, but the engineering underneath 간지사주 is not. The hard problem here is twofold: turning an open-ended, generative experience into something that loads instantly across a sprawling mobile surface, and bolting a real-money economy onto a consumer app where every double-charge or failed refund is a trust failure.

The approach treats the payment layer as the part that must never be wrong. Toss Payments sits behind explicit idempotency checks, a coin-credit ledger, an end-to-end refund workflow, and a settlement cron — so the same request can't be billed twice and reversals reconcile deterministically, separate from the probabilistic AI output users actually read. That AI layer is wrapped in its own discipline: cost tracking, feature telemetry, and fallback patterns for when generation degrades.

What makes it engineering-notable is the grounding move. Rather than letting the model freely invent, narrative generation retrieves from a Korean classical-text RAG corpus and cites its sources — the same instinct that runs through this builder's regulated-vertical work, applied to a consumer SaaS. The breadth (daily fortune, saju, compatibility, dream reading, date selection) is prerendered, and scheduled web push keeps the loop running without a backend babysitter.

Highlights

  • 256+ page mobile-first (daily fortune, saju, compatibility, dream reading, date selection), prerendered for instant load
  • Toss payments + idempotency checks + coin credits + refund workflow + settlement cron
  • OpenAI narrative generation + cost tracking / feature telemetry + fallback patterns
  • Korean classical-corpus RAG retrieval (cited grounding), 6×/day web push

Metrics

1,075
commits
256+
pages
53
migrations
8
cron jobs

Tech stack

Next.js 16React 19TypeScriptSupabaseOpenAI APIToss PaymentsTailwindPlaywrightVitest