Mario's Pizza

Mario's Pizza needed a solution to handle orders 24/7 without hiring overnight staff. The challenge was building an AI agent that could handle natural conversation in both English and Spanish. ARCHITECTURE Vonage Voice API handles inbound calls and DTMF detection. Claude AI takes the order through natural conversation. n8n orchestrates the workflow and integrations. Airtable stores orders and manages the menu. USER FLOW A customer calls the ordering line, the system plays a greeting in both languages, the caller presses 1 for English or 2 for Spanish, the Claude AI agent takes the order naturally, the system confirms items and total, the order is saved to Airtable, an SMS confirmation is sent, and the restaurant is notified via webhook. KEY FEATURES Natural conversation handles menu questions, modifications like extra cheese, upselling for drinks, and payment collection. Intelligent order processing validates items against the menu, calculates pricing automatically, handles special requests, and prevents duplicate orders. Multi-language support switches seamlessly between English and Spanish with consistent pricing and culturally appropriate responses. TECHNICAL CHALLENGES AND SOLUTIONS Call quality issues from background noise and accents were addressed with audio processing in n8n, transcription tolerance in Claude, and multiple confirmation attempts. Order accuracy concerns were solved with explicit confirmation after each item, a price recap before payment, and the ability to restart anytime. Real-time menu updates were handled by treating Airtable as the single source of truth, with n8n polling for changes and cache invalidation on update. RESULTS AND METRICS Before this system, order taking was manual and limited to business hours, with missed orders during peak times and staff burnout from repetitive calls. After launch, the restaurant had 24/7 availability, consistent order quality, and zero missed orders, freeing staff to focus on fulfillment. The system now processes over 200 orders per month with a 2-minute average call time, 98% accuracy, and 24/7 uptime. COST BREAKDOWN Vonage API runs about $100/month, Claude API about $150/month, the n8n server (Raspberry Pi) about $30/month, with Cloudflare and Airtable free — a total of roughly $280/month. That replaces the cost of one full-time employee (about $3,000/month), a 10x payback. LESSONS LEARNED Start simple with basic DTMF before adding natural language. Iterate with real users since they catch edge cases. Monitor everything including call quality and order accuracy. Keep a manual fallback available. Plan the architecture for 10x growth from day one. FUTURE ENHANCEMENTS Integration with delivery platforms like DoorDash and UberEats, time-of-day menu customization, loyalty program integration, an advanced analytics dashboard, and multi-location support.