Skip to content

replit.md

Latest commit

 

History

History
108 lines (92 loc) · 7.54 KB

replit.md

File metadata and controls

108 lines (92 loc) · 7.54 KB
Raw

Overview

This project is a pnpm workspace monorepo using TypeScript, focused on developing a Korean-Japanese dating app called "Lito" and its supporting services.

The Lito app (artifacts/lito) is an Expo (React Native) mobile application designed to connect users from Korea and Japan for dating and language exchange. It features a sophisticated matching system, real-time chat with translation capabilities, and a personalized user experience. Key capabilities include user authentication, profile management, a swipe-based discovery feed, real-time chat with message persistence, object storage for photos, and a contact blocking feature. The app aims to facilitate cultural connection and language learning.

Complementing the app is the Lito Admin dashboard (artifacts/admin), a React-based data visualization tool for trust & safety moderation, and an Express-based API Server (artifacts/api-server) providing backend services, including authentication, user management, chat, AI-powered features (matching, coaching, translation), and legal content serving.

The project's ambition is to create a leading platform for cross-cultural dating with robust AI integration and a strong focus on user safety and experience.

User Preferences

I want iterative development. Do not make changes to chat/[id].tsx. Do not reintroduce pronunciation features. Translation must remain free for all plans; never paywall it.

System Architecture

The project is structured as a pnpm workspace monorepo using Node.js 24 and TypeScript 5.9.

UI/UX Decisions (Lito App)

  • Theme: White background, rose/pink (#D85870) accent, dark charcoal text.
  • App Icons: Two overlapping hearts (coral pink + golden yellow) on a white background for icon.png (1024x1024) and splash.png (9:16).
  • Key Screens: Onboarding (3 slides), Login, Profile Setup, Discover (swipe cards), Matches, Chats, Chat Detail, Profile, Settings, Dating Style Diagnosis.
  • Component Design: Custom components like Button, CompatibilityChip, CountryFlag, ProfileImage, LitoMark, and a layered TrustBadge system.
  • Animations:
    • Chat bubble animations: fade + slide-up on entry (200ms), translation toggle crossfade (220ms), AI quick reply panel slide-up (250ms).
    • Discover card animations: fade + slide on entry (260ms), LIKE/PASS stamps with spring animation (scale 0.72→1), directional color overlays (rose tint for LIKE, grey tint for PASS), enhanced bottom gradient, and top scrim for readability.
  • Chat List UI: Filter tabs (All/Unread/Requests), unread badge on avatar, bold font + rose color for unread timestamps, improved time formatting.
  • Discover Cards: Language study badge (📚 pill, green glass), interests, and match reason chips.
  • Matches Screen: TrustBadge, trust dot overlay, study badge, AI ice breaker suggestion card, last active status.
  • Profile Screen: Section for cultural connection goals (KR↔JP flag bridge, language levels).
  • Lifestyle Filters: Smoking/Drinking habits with UI for selection chips and display, filtering options in Discover.

Technical Implementations (Lito App)

  • Monetization:
    • 3 plans: Free (20 likes/day, 3 picks), Plus (unlimited likes + boost), Premium (see who liked + AI coach).
    • Consumables: Boost, Direct Intro, City Pass, AI Review.
    • Entitlement system (isEntitled/useEntitlement) to check plan membership.
    • Mock billing system for future integration with RevenueCat/App Store.
    • Paywall screen for plan comparison and upgrades.
  • AI Matching (Heuristic):
    • computeCompatibility(): Scores based on intent, interests, cultural fit, conversation style, meeting feasibility.
    • generateChemistryPicks(): Daily ranked picks (3 for free, 10 for Plus/Premium).
    • generateProfileSuggestions(): Template-based intro/bio improvements.
    • generateOpeners(): Contextual conversation starters based on shared interests and country.
    • generateChemistryCard(): Deterministic dating-type card (4 types).
  • Viral / Referral:
    • generateReferralCode(): Deterministic userId prefix + random suffix.
    • Reward system: boost on signup, direct intro on first match.
    • applyReferralCode(): Validates and records referral attribution.
    • Referral screen for code/link sharing, reward claiming, and stats.
  • Analytics: trackEvent() facade covering 24 events across monetization, AI, and viral categories, with console logging in development.
  • Authentication: Email/password registration/login with JWT (30-day expiry). Social login support for Google, Apple, Kakao, and LINE (server-side OAuth). Deep linking for auth callbacks.
  • Real-time Chat: WebSocket-based (ws package) with JWT authentication, room-based broadcasting, message persistence in DB, auto-reconnection, and HTTP fallback.
  • Object Storage: Google Cloud Storage (GCS) based for photo uploads, utilizing presigned PUT URLs for direct client-to-GCS uploads, with server-side serving of objects.
  • Discovery: DB-driven discovery with swipe_passes table, enabling actual user interaction tracking (like/pass/match). AI personas auto-match, demo users are in-memory, and unauthenticated guests see demo/AI users.
  • Contact Blocking: expo-contacts and expo-crypto are used to hash contact numbers client-side (SHA-256) before sending to the server (contact_block_hashes table), ensuring two-way blocking in the Discover feed.

System Design Choices (API Server)

  • API Framework: Express 5.
  • Database: PostgreSQL with Drizzle ORM.
  • Validation: Zod (zod/v4) and drizzle-zod.
  • API Codegen: Orval from OpenAPI spec.
  • Build: esbuild (CJS bundle).
  • Logging: Pino.
  • Architecture: Modular 3-layer design (Router → Service → Repository).
    • Router: Handles I/O validation only.
    • Service: Contains business logic and LLM calls.
    • Repository: Manages DB queries only.
  • Interest Signal System v4:
    • conversation_participants: Manages chat memberships.
    • feature.extractor: Extracts conversation features from messages.
    • interest.repository: Handles interest_snapshots and latest_interest_snapshots DB I/O.
    • llm.circuit: Implements an LLM circuit breaker for resilience.
    • interest.worker: Asynchronous worker for analyzing interest signals with debouncing.
    • interest.service: Computes PRS (Persona-Relationship Score) using deterministic and LLM-based logic.
    • ws.ts: WebSocket gateway for real-time communication, JWT authentication, room authorization, and interest signal push.
  • AI Coaching/Language Features: /ai/coach, /ai/suggest-reply, /ai/translate, /ai/persona, /ai/conversation-starter, /ai/generate-profile-photo.

External Dependencies

  • Monorepo Tool: pnpm workspaces
  • Package Manager: pnpm
  • API Framework: Express 5
  • Database: PostgreSQL
  • ORM: Drizzle ORM
  • Validation: Zod
  • API Codegen: Orval
  • Build Tool: esbuild
  • Logging: Pino
  • Mobile Framework: Expo (React Native), Expo Router
  • State Management (Mobile): AsyncStorage
  • Contact Management (Mobile): expo-contacts
  • Cryptography (Mobile): expo-crypto
  • Real-time Communication: ws package
  • Object Storage: Google Cloud Storage (GCS)
  • UI Framework (Admin): React
  • Routing (Admin): Wouter
  • Data Fetching (Admin): TanStack Query
  • Styling (Admin): Tailwind CSS
  • OAuth Providers: Google, Apple (via expo-apple-authentication), Kakao, LINE
  • AI/LLM: OpenAI (via src/infra/openai)
  • Analytics: PostHog/Amplitude (facade present, integration points defined)