Meridian
A cutting-edge real-time ride-hailing simulation platform demonstrating event-driven architecture, live data streaming, and AI-powered services across multiple cities.
System Architecture
Event-driven microservices communicating through Apache Kafka, enhanced with AI for intelligent pricing and natural language interactions.
Loosely-coupled services communicate asynchronously via Kafka topics (ride, taxi), enabling independent scaling and fault isolation
Ride Journey Flow
Chat Request
User chats with MeridianAI in natural language. The ride-api uses Claude AI to sanitize input and extract pickup/dropoff locations, then geocodes them via Mapbox.
Match & Assign
Orchestrator consumes from ride topic, finds nearest available taxi within 2.5km, or generates new taxis 0.8-2km away. Publishes assignments with routes from Mapbox Directions API.
Live Movement
Simulator replays taxi movement along actual Mapbox routes, streaming position updates every few seconds to taxi topic. Frontend receives updates via WebSocket bridge.
Smart Billing
Billing service uses Claude AI to extract real taxi rates for the location, cached in Redis for 24h. Calculates fare in local currency (GBP, EUR, RON, INR), persists to PostgreSQL.
Key Features
⚡ Event-Driven Design
Kafka topics (ride, taxi) enable asynchronous communication between services. FastAPI ride-api bridges WebSocket connections to Kafka consumers, streaming real-time updates to the React frontend as events flow through the system.
🤖 Dual AI Integration
MeridianAI chat uses Claude via DigitalOcean AI Inference API to parse natural language ride requests and extract locations. Billing service queries Claude/GPT-4o to fetch real-world taxi rates for any city, with Redis caching (24h TTL) to minimize API calls.
🗺️ Mapbox Integration
Geocoding API converts locations to coordinates. Reverse Geocoding detects city/country from coordinates for locale-aware names (via Faker). Directions API provides routes that the simulator replays point-by-point for realistic movement.
🌍 Global Realism
System adapts to any location worldwide. Orchestrator intelligently generates taxis 0.8-2km away (never on the same street) when none are nearby. Multi-currency billing (GBP, EUR, RON, INR) with location-specific rates. Locale-aware name generation produces authentic names for each city using Faker locales.
📊 Data Persistence
NeonDB Serverless PostgreSQL stores ride completions with billing details for analytics. Redis caches taxi rates by city. Database schema includes ride status tracking, location context, and fare breakdowns with unique constraints on (user, taxi, timestamp).
🐳 Container Orchestration
All services run in Docker containers orchestrated via docker-compose. Caddy reverse proxy handles HTTPS with automatic certificate management (Cloudflare DNS). Kafka UI provides topic inspection. Supports both local dev and production deployments.
Technology Stack
Demo Video
Watch Meridian in action: matching, routing, simulation and AI-powered billing.