Oxytrip Travel Agency

Oxytrip is a UAE-based travel agency serving customers across the Persian Gulf, a region where a single conversation might move between Arabic, Farsi, Urdu, Hindi, Russian, and English, sometimes within the same thread.

Like most growing travel agencies, Oxytrip was running into a familiar wall: demand for support was constant, but staffing a live team around the clock, in every language their customers spoke, wasn't realistic without a significant increase in headcount.

We built an AI Travel Assistant to solve that, not a simple chatbot bolted onto a website, but a full support and sales layer running across the channels Oxytrip's customers actually use.

The Problem in Five Parts

Before designing anything, we mapped exactly what was creating pressure on Oxytrip's support operation:

• Round-the-clock demand. Travel questions don't wait for business hours. Staffing live agents 24/7 is expensive, and the alternative — slow responses outside working hours — costs bookings.
• A genuinely multilingual customer base. Arabic, Farsi, Urdu, Hindi, Russian, and English speakers each expect to be served in their own language, not routed to whichever agent happens to be available.
• High-volume, low-value questions. Baggage allowances, check-in procedures, visa basics, refund policy, these questions are simple, but there are a lot of them, and they were consuming time that could go toward actual sales conversations.
• Brand and compliance risk. A wrong answer about visa requirements, refund eligibility, or pricing isn't a minor inconvenience in travel, it can cost a customer their trip. A generic chatbot willing to guess is a liability, not a convenience.
• WhatsApp-first customers. The Gulf market runs on WhatsApp. But mixing bot responses and human replies on the same number, without a clear system, creates confusion,  customers don't know if they're talking to a person or a script, and staff don't know when to step in.

What We Built

The AI Travel Assistant is a single, lightweight system that meets customers on Oxytrip's website widget and on WhatsApp, answers from a curated knowledge base, runs live flight searches directly against Oxytrip's own booking system, and hands off anything involving money, documents, or uncertainty to a human agent, through a dedicated Telegram desk, translating the conversation in both directions.

The core design principle was simple: the assistant only speaks from approved information. It does not improvise on facts that could cost a customer money or damage the agency's credibility.

Every message follows the same path: the system reads what the customer wrote, classifies the intent and language, decides whether it can answer safely, and either responds directly or escalates to a person, logging the full exchange either way.

On its own, the assistant can:

• Answer frequently asked questions — baggage, check-in, visas, documents, payments, cancellations, hotels — instantly, in whatever language the customer is writing in, drawn entirely from Oxytrip's approved knowledge base.
• Search live flights, resolving city names to the correct airports, querying Oxytrip's booking system in real time, and presenting up to five real fares directly in the conversation, complete with smart handling of flexible dates and nearby airport alternatives.
• Qualify and hand off, collecting the details of a trip, then routing a complete, sales-ready summary to a human agent to close the booking and take payment.
• Everything outside a narrow set of general topics — flight duration, weather, packing, local customs, and ordinary conversational back-and-forth — must match the approved knowledge base or the conversation escalates. There's no middle ground where the system is guessing.

What This Looks Like in an Actual Conversation

Here's a real interaction pattern the system handles on WhatsApp, from a vague opening question to a completed handoff:

Customer: "Any flights to Karachi next week?"

Assistant: Recognizes this as a flight search, notices the origin, exact date, and passenger count are missing, and asks for each. one at a time, in the customer's own language.

Customer: "From Dubai, the 20th, 2 adults."

Assistant: Resolves Dubai to DXB and Karachi to KHI, runs a live search, and presents the top five real fares with times, prices, and stops.

Customer: "Book the first one."

Assistant: Refreshes the price to make sure it's still current, then connects a human agent with a complete booking brief — route, date, passenger count, and current fare — ready to take payment.

What looks like a simple exchange to the customer is, behind the scenes, resolving ambiguous inputs, querying a live booking API, caching results to avoid redundant lookups, and handing off with enough context that the human agent never has to ask the customer to repeat anything.

The system also handles the messier real-world cases: vague date requests like "after Eid" get resolved to actual dates instead of the conversation looping. A country-level request like "flights to Pakistan" triggers a city picker instead of a failed search. If there's no available inventory, the assistant suggests a nearby airport once, then hands off, it doesn't keep guessing.

Protecting the Brand While Doing the Work

The part of this project we're most proud of isn't the flight search, it's the discipline built into what the assistant is allowed to say.

By answering only from approved content and escalating anything involving bookings, payments, visas, or genuine uncertainty, the system is structurally prevented from giving a confident wrong answer on the exact topics — visas, refunds, pricing — where a mistake actually costs the business trust and money.

Oxytrip's team also has full visibility and control: every conversation, every escalation, and every confidence score is logged and reviewable. Staff can edit the knowledge base directly, tune how cautious the system is, and turn it on or off per channel, no engineering request required.

• 24/7 autonomous coverage across both the website and WhatsApp, no gaps in availability, no added headcount.
• 8+ languages served from a single English-speaking desk. Translation runs in both directions, so one team can effectively support a customer base speaking Arabic, Farsi, Urdu, Hindi, Russian, and English.
• Live fare search built directly into the conversation up to five real fares presented in-chat, moving customers from "what's the price?" to a bookable option without ever leaving WhatsApp.

Beyond the numbers, the practical shift for Oxytrip's team has been what gets automatically deflected versus what reaches a person. Routine questions — the ones that used to consume agent time without generating revenue — are now handled instantly and around the clock. What reaches a human is, by design, the conversation that's actually ready to close.

What's Next

This system was built to expand, not stay fixed in place. The roadmap includes:

• Extending the same collect-and-handoff pattern to hotel and package search alongside flights.
• In-chat booking and payment for straightforward itineraries, keeping human agents focused on the complex trips that need them.
• An analytics dashboard surfacing deflection rate, top routes, and lead conversion for management visibility.
• A move to a managed database as conversation volume grows beyond what the current lightweight setup was built for.

The Bigger Point

This project is a good example of what we mean when we talk about the difference between a chatbot and a real operational system.

A chatbot would have answered questions. What Oxytrip needed — and what we built — actively searches live inventory, qualifies leads, protects the brand from costly mistakes, and hands off to a person at exactly the right moment, in whatever language the customer walked in speaking.

If you're running a business where support volume, language coverage, or after-hours demand is creating the same kind of pressure Oxytrip was facing, that's exactly the kind of problem this approach is built to solve.

📩 If you'd like to talk through what a system like this could look like for your business, contact us.