Likeyou AI
AI-powered matchmaking platform that connects people through intelligent reasoning, not just swipes.
Overview
The Problem
Dating apps reduce connection to superficial swiping. Users spend hours on matches that go nowhere because compatibility is judged by photos alone. The result: swipe fatigue, shallow conversations, and wasted time for everyone.
The Solution
Likeyou uses AI to match based on personality and interests, not looks. Describe your ideal connection in natural language - "someone who loves philosophy and late-night conversations" - and let the AI find compatible profiles.
Architecture
┌─────────────────────────────────────────────────────────────────┐
│ FRONTEND │
│ Next.js 14 (App Router) + TypeScript + Tailwind │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ API ROUTES │
│ ┌─────────────────┐ ┌─────────────────┐ │
│ │ /api/discover │ │ /api/chat │ │
│ │ (AI Matching) │ │ (Real-time) │ │
│ └─────────────────┘ └─────────────────┘ │
└─────────────────────────────────────────────────────────────────┘
│
┌───────────────┼───────────────┐
▼ ▼ ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ LLM API │ │ SUPABASE │ │ SUPABASE │
│ (AI Reasoning) │ │ (PostgreSQL) │ │ (Realtime) │
│ │ │ Users/Profiles │ │ Chat/Presence │
└─────────────────┘ └─────────────────┘ └─────────────────┘AI Layer
LLM processes natural language queries and computes compatibility scores.
Data Layer
Supabase handles auth, user profiles, and connection state.
Real-time Layer
Supabase Realtime powers chat, typing indicators, and presence.
Tech Stack
Every tool was chosen for a specific reason:
LLM API
Chosen over OpenAI for better reasoning on nuanced personality matching. The 1M context window allows processing multiple profiles in a single inference call.
Supabase
PostgreSQL with built-in auth, real-time subscriptions, and row-level security. One platform replaces Firebase + custom auth + WebSocket server.
Next.js 14
App Router for server components, reducing client bundle size. API routes colocate backend logic with frontend, simplifying deployment.
TypeScript
Type safety across the stack - from API responses to database queries. Catches integration bugs before they reach production.
How It Works
1. AI Vibe Match
Describe your ideal connection in natural language. The AI searches profiles for compatible matches.
2. Browse Profiles
View AI-calculated compatibility scores. Each card shows occupation, bio, and match percentage.
3. Manage Connections
Track sent requests, received requests, and established connections in one place.
4. Real-time Chat
Message matches with text, images, voice notes, and video calls.
Challenges
AI Response Latency
Problem: LLM API calls take 2-5 seconds. Users expect instant search.
Solution: Implemented streaming responses with loading states. Pre-computed compatibility embeddings for frequently searched traits.
Real-time Chat at Scale
Problem: Managing WebSocket connections and message ordering.
Solution: Leveraged Supabase Realtime instead of custom WebSockets. Optimistic UI updates with server reconciliation for perceived speed.
Matching Algorithm Accuracy
Problem: Early prompts returned irrelevant matches.
Solution: Iterated on prompt engineering. Added structured output format with explicit scoring rubric. Included negative examples in system prompt.
Results
What Works Today
- • AI-powered profile discovery
- • Natural language search queries
- • Full connection request flow
- • Real-time chat with media support
- • User authentication and profiles
Planned Next
- • Video call integration
- • Enhanced AI personality analysis
- • Icebreaker question suggestions
- • Mobile app (React Native)
- • Premium matching tiers