WhatsApp AI agent and customer-service orchestration for a travel agency
The project
Meraj Travel, a Brussels-based travel agency, handled all its client requests through WhatsApp, with manual and sequential processing. Messages, quotation requests, modifications, logistics questions, file tracking, arrived on a single number, without prior qualification or clear distribution among advisors. The result was irregular response times, messages lost during peak periods, and a team forced to constantly juggle WhatsApp with its internal tools.
What I designed and led
I designed a three-tier conversational system, following a simple principle, the AI qualifies, Chatwoot orchestrates, the advisor finalizes. Chatwoot centralizes all WhatsApp conversations in a multi-agent interface, with a persistent customer history accessible to the whole team. An AI agent layer, powered by n8n workflows, intercepts messages and passes them to an LLM enriched with business context, offers, FAQs, conditions, customer files, generating either an autonomous response or a qualified request with a routing suggestion. Finally, a router automatically assigns each request to the right advisor or team based on its type, reservations, modifications, support, quotations, with purely informational conversations closed by the AI.
The choices that make the difference
A single entry point that eliminates fragmentation across personal phones. A progressive handoff logic, where the AI only handles scoped domains, FAQs, hours, general information. Business context injected into every LLM call. Native traceability for audit and continuous prompt improvement. And an AI deliberately prevented from making any commercial commitment, with confirmations remaining the responsibility of a human advisor.
What it shows
Instant first responses, automatic qualification of requests and clear distribution across the team, freeing advisors for high-value conversations while the AI agent absorbs repetitive inquiries. This project shows my ability to design a genuinely useful AI agent, integrated with existing tools and scoped to stay reliable in production.