AI Restaurant Copilot
Voice AI Agent for Restaurant Operations
<200ms
Webhook Latency
92%
Fuzzy-Match Accuracy
O(1)
Upsell Retrieval
Executive Overview
Engineered for scale and business impact, the Restaurant AI Copilot is an end-to-end ecosystem that transforms raw restaurant data into actionable revenue strategies. Leveraging React, Vercel Serverless Functions, and Supabase, the platform provides real-time margin analytics — factoring in complex delivery platform commissions. Its crown jewel is a custom NLP-driven Voice Copilot built on Twilio, capable of understanding Hinglish, parsing modifiers, resolving ambiguities via Levenshtein distance algorithms, and autonomously executing data-backed cross-sells.
The Problem
Restaurants struggle with phone-order overload, missed calls, and slow response times during peak hours.
The Solution
Built an AI voice ordering agent paired with a revenue analytics dashboard to handle customer calls instantly.
Core Capabilities
Voice Order Processing
Handles customer calls naturally to take food orders.
Live Analytics Dashboard
Visualizes incoming orders and revenue in real-time.
Natural Language Understanding
Extracts items, quantities, and special instructions.
Revenue Intelligence
Channel-aware profitability with Hidden Star & Unprofitable SKU detection.
System Architecture
Architecture Flow
How data moves through the system, from user interaction to backend processing and response generation.
Customer Call
Customer dials the restaurant number. Twilio triggers a Webhook to the Vercel API.
Interface Gallery
Command Center / Analytics Dashboard
Technical Deep Dive
voiceEngine.ts — Custom NLP
Highly optimized Regex and Fuzzy-Matching intent engine using Levenshtein Distance to handle typos, accents, and poor transcriptions. Categorizes intents (Greeting, Order, Modify, Cancel) with modifier extraction loops.
aiEngine.ts — Data Science in TS
Builds a mathematical Co-Occurrence Matrix mapping Item A × Item B intersections. Uses Bayesian-like probability for cross-sells. Sales Velocity Engine uses exponential decay lambda for micro-trend spotting.
Stateless Twilio Hydration
Every Vercel API execution pulls current call state from Supabase (hydrateSessionFromCallLog), processes speech, determines next conversation node, updates state in PostgreSQL, returns TwiML.