A D2C mobile commerce that sells AI-generated saju reports as products. Payments, alerts, ad-conversion tracking and the operating back office all run inside one system.
Selling a saju report as a product means content generation, payments, alerts and ad conversions all have to move in lockstep.
Showing the AI report isn't the end. A guest has to finish checkout in one go, a refund pressed in Toss has to sync back to the admin, KakaoTalk has to fire purchase and result notifications automatically, and the ad pixel has to register the conversion accurately.
The operator already knew that farming each of those pieces out to a different vendor would leave at least one gap where nobody owns the outcome.
We stitched order through report-retrieval into a single automation pipeline, and designed an admin the operations team can run daily without engineers.
Claude and Gemini generate the saju report. A guest order automatically provisions a user, and a later Kakao login fuses the accounts. Toss Payments handles the full purchase / refund / cancel flow — the admin's refund button calls the Toss API for you.
Ad conversion is dual-tracked via the Meta Pixel and Conversions API, and surface metrics like 'lifetime purchases' or 'CSAT' are editable from the admin in real time, so the marketing team doesn't wait on engineering.
App Router-based full-stack. User-facing screens and the admin back office live in one codebase.
Users, orders, products, templates, media and site settings share one schema, accessed type-safely to cut runtime errors.
Saju reports branch across two models. Prompts edit live in the admin so analysis tone can be tuned during operations.
Full payment / refund integration, Kakao OAuth login, automated KakaoTalk alerts — one purchase cycle, one place.
Client and server dual-track conversion events. Data still arrives under iOS tracking restrictions.
Elastic Beanstalk + RDS + CloudFront + S3. main/develop branches deploy automatically into separate staging and production.