A comprehensive massage and SPA knowledge encyclopedia app built with Expo + React Native, deployed on Vercel with Supabase PostgreSQL.

Live: https://spa.rxcloud.group

• 30 knowledge articles across 6 categories
  • 按摩基础知识 (Basics)
  • 常见按摩技法 (Techniques)
  • 精油与香薰 (Essential Oils)
  • 穴位与经络 (Acupoints)
  • SPA 礼仪与体验 (SPA Etiquette)
  • 男士 SPA 消费指南 (Men's SPA Guide)

• 症状速配 (Symptom Match) - 3-step massage recommendation based on symptoms
• 穴位计时器 (Acupoint Timer) - Guided massage with rhythm and timing
• 每日挑战 (Daily Challenge) - 30-day wellness challenge with streak tracking
• AI 顾问 (AI Advisor) - Intelligent Q&A for massage and wellness
• 知识问答 (Knowledge Quiz) - Interactive quiz to test your knowledge

• Full-text search with hot search tags
• Favorites with local persistence (AsyncStorage)
• Responsive layout (mobile, tablet, desktop)
• Dark mode support
• Sub-heading detection for structured content rendering
• Related article recommendations

The AI Advisor uses a RAG-like pattern: knowledge articles are indexed in the system prompt, the LLM references article IDs in responses, and the client renders them as clickable deep links.

Key design decisions:

• Knowledge Index in System Prompt — 30 article IDs + acupoint/oil quick reference injected as context, avoiding vector DB complexity
• 10-message Sliding Window — balances conversation continuity vs. token cost
• Article Link Parsing — LLM outputs [title](article:ID) format, client-side ChatBubble parses into navigable Pressable components
• Stateless Server — no server-side chat storage; client owns persistence via AsyncStorage (50 msg limit)
• Thinking Budget — Gemini 2.5 Flash with 128-token thinking budget for reasoning

Request lifecycle:

1. User sends query → useChat hook extracts last 10 messages as context
2. tRPC mutation posts to advisor.chat (superjson serialized, Zod validated)
3. Server prepends system prompt (knowledge index + acupoint/oil reference + rules)
4. Forge API call → Gemini 2.5 Flash processes with thinking budget
5. Response with [article](article:ID) links returned to client
6. ChatBubble renders clickable article links → user navigates to full knowledge detail

cd massage_spa_guide
pnpm install
pnpm dev

massage_spa_guide/
  app/                    # Expo Router screens
    (tabs)/               # Bottom tab navigation
      index.tsx           # Home with tools & categories
      search.tsx          # Full-text search
      advisor.tsx         # AI chat advisor
      favorites.tsx       # Saved articles
    category/[id]         # Category listing
    knowledge/[id]        # Article detail
    tools/                # Interactive tools
      symptom-match.tsx   # Symptom-based recommendations
      acupoint-timer.tsx  # Guided massage timer
      daily-challenge.tsx # Daily wellness challenge
      challenge-stats.tsx # Challenge statistics
      mens-guide.tsx      # Men's SPA consumer guide
      mens-quiz.tsx       # Consumer knowledge quiz
  api/                    # Vercel serverless entry point
  data/                   # Static knowledge content (JSON)
    knowledge.json        # All articles & categories
    symptom-matrix.json   # Symptom matching rules
    routines.json         # Massage routines
    challenges.json       # Daily challenges
  server/                 # Express + tRPC backend
  drizzle/                # Database schema & migrations
  components/             # Reusable UI components
  lib/                    # Client utilities (tRPC, theme)
  hooks/                  # React hooks

The app is deployed on Vercel with the following setup:

• Frontend: Expo web export (npx expo export --platform web) served as static files
• API: Express app as Vercel serverless function via api/index.ts
• Database: Supabase PostgreSQL (table: spa_users)

MIT