Korean Saju AI - AI-Powered Korean Fortune Reading App

Built with Next.js/TypeScript frontend + FastAPI backend; stateless architecture, no persisted user data.
Cut LLM generation cost ~75% by restructuring prompts for cache-eligibility and replacing a static rulebook with Python-precomputed, chart-specific relationships (~18% fewer tokens/call, zero quality loss).
Added Redis-backed rate limiting and abuse protection.
Shipped full EN/KO/TC localization and a custom design system.

Tags:

Next.jsTypeScriptFastAPIPythonLLMPrompt EngineeringRedisRate Limitingi18nDesign SystemWeb app

Screenshots: Korean Saju AI