Travel · client project · AI automation
An AI assistant that books travel end to end, safely

The problem
A travel agency's booking flow was fully manual: every search, quote, and confirmation went through a human, limiting hours and speed.
What I built
- A WhatsApp assistant, live on the agency's business number, running the full booking conversation in natural, human-sounding language.
- A Dify chatflow handling intent, context, and guardrails.
- A custom MCP server connecting the AI to the booking system and database. The AI drives search, pricing, and booking up to payment.
- Ticketing and payment deliberately gated behind protected internal endpoints. The AI can never complete a financial action on its own.
- A human-in-the-loop escalation path: when the assistant does not know an answer, it notifies the owner privately instead of guessing. The owner supplies the reply, the assistant picks the conversation back up, and the owner curates the knowledge base through an admin panel.
Architecture
WhatsAppDify chatflowMCP serverPermission gateSearch · price · book
Ticketing & payment: internal endpoints only· fails safely
The outcome
A working end-to-end AI booking flow, live on WhatsApp, with safe action-gating and honest escalation: the assistant does the work, the guardrails keep it inside its lane, and it never bluffs a customer.
What this proves
Real AI automation with permission-safe action-taking and a human escalation path. This was my second shipped MCP implementation.
Stack
WhatsAppNode.jsTypeScriptDifyMCPPostgreSQLGDS integration
Want something like this?
Tell me your most repetitive task, the audit is free.
Book a free AI automation audit