Circles – AI-Powered Location-Based Social Network

Executive Summary

Circles is a location-based social network powered by AI agents that connects people around shared objectives and real-world intentions. Users create "Circles" – intention bubbles represented by personalized AI agent personas – that collide in the real world to facilitate authentic, high-quality connections.

The Problem

Traditional location-based social apps suffer from:

• Passive Discovery: Users endlessly browse similar profiles without understanding intent or compatibility
• Low Match Quality: No real assessment of whether two people will actually enjoy connecting
• Compromised Privacy: Too much personal data exposed during discovery
• Decision Fatigue: Overwhelming number of options without quality criteria

The Solution

Circles uses AI agent personas built from existing digital traces (social media, chats, events) to:

1. Create Circles: Specify objectives (e.g., "Play tennis this evening in Palermo"), radius, and duration
2. Detect Collisions: When circles overlap geographically, agents evaluate compatibility
3. Simulate Interactions: Agents conduct brief simulated conversations to assess if connection is truly "worth it"
4. Present Matches: Strong alignments open direct chats with agent-generated explanations

Key Features

• Personalized AI Agents: Each user has an agent that learns their interests, communication style, and boundaries
• Semantic Matching: Uses embeddings to intelligently align objectives and interests
• Multiple Circles: Users can maintain multiple active circles simultaneously for different purposes
• Privacy-Preserving: Agents explain matches without exposing sensitive personal data
• Agent Simulations: Automated evaluations of connection potential before contacting users

Technology Stack

Backend

• Framework: Express.js (Node.js + TypeScript)
• Database: PostgreSQL with PostGIS extension
• ORM: Prisma (type-safe, automatic migrations)
• Validation: Zod (runtime validation with TypeScript type inference)
• Authentication: Passport.js + JWT + bcrypt
• Message Queue: BullMQ (for async agent processing)
• Cache: Redis (geospatial indexes and session caching)

Mobile Frontend

• Framework: Flutter (Dart)
• Features: Background location, push notifications, real-time chat with WebSockets

AI Features

• LLM Models: AWS Bedrock for agent simulations
• Embeddings: Semantic vector processing of objectives and interests
• Profile Building: Multi-source data consolidation and analysis

Core User Flow

1. Onboarding: User registers, optionally connects data sources (social media), completes questionnaire
2. Agent Creation: Backend builds an AI persona from collected data
3. Create Circle: User specifies objective, radius, duration
4. Collision Detection: System identifies geographically overlapping circles
5. Compatibility Evaluation:
   • Semantic matching: Compare objectives and interests
   • Agent simulation: Agents converse to evaluate "worth it score"
6. Match Presentation: Strong matches open direct chat; soft matches request consent
7. Chat: 1:1 communication with AI-generated suggestions

Success Metrics

• Match Quality (measured by engagement and user feedback)
• Privacy: Zero exposure of non-consented sensitive data
• Adoption: Active users creating multiple circles
• Safety: Detection and prevention of malicious behavior

Key Differentiators

1. Real Agents: Not just algorithmic filtering – agents simulate actual conversations
2. Privacy by Design: Agents explain connections without exposing sensitive data
3. Real Intention: Circles are based on concrete objectives, not just interests
4. Quality Over Quantity: Few high-quality matches instead of many superficial ones

Why Circles Matters

• Consumer AI Track: Demonstrates practical application of AI agents to solve real social friction
• Innovation: Uses multi-agent simulations for authentic human connection discovery
• Technical Depth: Combines geospatial indexing, semantic embeddings, and LLM agent orchestration
• Privacy-First: Shows how AI can enhance experiences while protecting user data

---

Circles is the result of a 36-hour hackathon. We're building the future of meaningful human connection.