ToteTaxi — Luxury Delivery Platform
Full-stack luxury delivery and mini-moving platform with an AI customer service agent, dynamic pricing, Stripe payments, and Onfleet dispatch.
Project Overview
ToteTaxi is a production logistics platform serving Manhattan to Hamptons, South Florida to NYC airports, and the NYC tri-state area. It handles four service types — Mini Moves (tiered packages), Standard Delivery, Specialty Items, and Airport Transfers — with a sophisticated dynamic pricing engine that accounts for geographic surcharges, weekend/peak premiums, same-day restrictions, and promotional discount codes.
The Challenge
A luxury delivery business needed a complete platform — booking, pricing, payment, dispatch, customer service — that could handle complex logistics across multiple service areas and service types.
The Solution
Built a full-stack platform with an AI customer service agent (LangGraph + Claude, 6 tools, SSE streaming), dynamic pricing engine, Stripe payment processing with atomic booking creation, Onfleet driver dispatch with real-time webhook tracking, and separate customer/staff portals with role-based access.
Technology Stack
Backend
Frontend
Ai
Integrations
Deployment
Key Features
AI agent with 6 tools: ZIP coverage, real-time pricing, date availability, booking lookups, hand-off to booking wizard with up to 22 pre-filled parameters
Dynamic pricing engine with geographic surcharges, weekend premiums, same-day restrictions, and promo codes
Dual authentication — httpOnly cookies for desktop, session fallback for mobile
Onfleet driver dispatch with real-time webhook status tracking
Stripe payment processing with atomic booking creation
Automated email confirmations with calendar invites
310+ backend tests via pytest, Playwright E2E tests
Business Impact
Complex multi-zone pricing with dynamic surcharges and restrictions
AI agent that accurately routes services, validates dates, and pre-fills booking forms
Concurrent SSE streaming alongside standard API traffic via gevent workers
Auth-scoped tool binding preventing LLM data access violations
Technical Achievements
Production LangGraph agent with adversarial-tested system prompt
Full security audit covering payment hardening, rate limiting, and LLM attack surface
310+ tests and Playwright E2E coverage
Gevent worker architecture for concurrent SSE + API traffic
Future Enhancements
Real-time driver tracking map for customers
Recurring delivery scheduling
Multi-language support
Technical Implementation
Built with Django 5.2 and Next.js 16. The AI agent uses LangGraph's ReAct pattern with Claude, streamed via SSE. The system prompt was iteratively refined through adversarial testing. Auth-scoped tool binding enforces data access boundaries at the code level. LangSmith provides full observability into agent behavior in production. Backend runs on Fly.io with gevent workers for concurrent streaming.