In this exclusive interview, Dr. Shakti Goel, Chief Architect and Data Scientist, and Upasana Gupta, Head of AI, Yatra Online, share with CXO Media how the company is using AI, data, and cloud technology to transform travel experiences.
From personalised recommendations to its GenAI assistant DIYA, the company is building smarter, more intuitive travel solutions.
How does Yatra Online create a personalised travel experience for its customers? What role does technology play in enhancing customer experience?
We have an enterprise data warehouse with data for multiple LOBs, such as flight and hotel, and for B2C, B2B, and B2E businesses. We store data for multiple subject areas. We use this data to identify the “likings” of the customer and create personalized options. If a customer is searching flights, we provide flight options that (s)he is most likely to book. We use Collaborative filtering and ML algorithms such as Hierarchical Recurrent Neural Networks to personalize search results. Our travelers can create personalized itineraries using our bot DIYA. The itinerary comes linked to a live inventory of flights and hotels. The customer experience is enhanced because now they can choose flights and hotels and estimate the cost of the trip. Soon, we will be launching multi-city and multi-country itineraries, and the customers will be able to book hotels in different cities, thus creating a complex dynamic package. For example,
“I want to go to Europe and visit 3-4 countries over a period of 7 days. I will leave from Delhi on May 15 with a two-year-old. I want hotels to be near the train station. The trip to other cities can be day trips or trips with an overnight stay. I want 3* or higher hotels.”
This will give you flight options from Delhi to Europe, suggest a detailed day-by-day itinerary, and recommend hotels suitable for a toddler and near the train station in various cities. It will also display a suggested price for the package. The itinerary will also suggest top restaurants, top markets, and top attractions to provide more travel inspiration. All this can be achieved in a single query from the traveler!
We have introduced personalization in our Expense Management Solution. Travel policies are customized and linked to the user filing the expenses. The user can file expenses using voice and without typing a single word! This makes the product friendly for physically challenged individuals – a leap in supporting social welfare.
We also have an email bot that responds to corporate queries on flight options. A lot of emails come from our travel partners informing us of flight changes (delays, cancellations, rescheduling). A personalized communication is created and sent to our retail customers.
Technology, whether AI-related or otherwise, has a significant role to play in enhancing customer experience. We heavily use the LLM-backed GenAI technology to enable our products with natural language. Modern-day technologies of React.js and Node.js, in conjunction with websockets and real-time streaming, are used to provide an interactive experience. In order to have maximum bang for the buck, we work with one-stop technologies such as Flutter, using which we can create a web application, an Android and iOS mobile application, a responsive mobile site, and a progressive web application (PWA). What’s more, the same app can be deployed on tablets as well. Our working principle is to solve business problems using technology, and not get enamored by technology itself.
What are the major features of DIYA, Yatra’s generative AI-powered travel assistant? What was the biggest technical challenge in building a multilingual AI agent?
DIYA, short for Digital Intelligent Yatra Application, is an ML and GenAI-powered travel concierge suite of applications, which are voice-enabled and work in 100+ languages and scripts. There is a version of DIYA for B2C customers and enterprise / corporate customers. DIYA is a multi-faceted application that allows for flight and hotel searches using natural language. Search can be made for flights and hotels in the same query. A complex travel itinerary driven by the traveler’s preferences can also be created and tied to the most relevant flight and hotel options. As mentioned above, a multi-city/multi-country itinerary can also be created, and hotels can be booked in different cities. DIYA allows a traveler to create dynamic packages.
As a concierge would do, DIYA allows users to cancel flights and hotels, determine the cancellation charges, download flight e- tickets and hotel vouchers, and download invoices. Our customers have asked us to provide details of upcoming bookings (flights, hotels, trains, buses), and DIYA does that. If the customer has any queries related to travel, Yatra services such as Prime, policies, and many others, then a GenAI-enabled and natural language-driven FAQ section is available.
One very interesting feature of DIYA is image analysis. Say, you are on a trip and are curious to know about a monument, sculpture, or painting. You could use Google Lens for it. But DIYA provides a far more enriching experience. It will tell you not only about the monument but also its history, any myths associated with it, architectural history, FAQs, and a lot more. It is like having a travel guide holding your hand and acting like a personal assistant.
We recently introduced a feature where a person can set the flight price in DIYA, and whenever the price drops, DIYA will alert the traveler. During the flight cancellation crisis faced by Indigo in December of 2025, DIYA provided a feature where travelers could check if their flight was cancelled and whether the refund was issued by the airlines. No human involvement was needed.
An AI Agent has the primary challenge to understand the intent of the user query. The classifier agent can sometimes get confused. For example, if the customer asks for booking details, is she want to see the past bookings or the future bookings? Making the agent multilingual has its own challenges. The LLM can hallucinate and change the language in which the conversation is happening, even though the user has not asked for the language switch.
A lot of people fail to understand that Agentic AI technology is a façade over underlying APIs and Data. If the API ecosystem is not stable or the data is not curated, then the Agentic AI technology will not work. What is Agentic AI? In a way, it is an orchestration of APIs. Furthermore, no technology will work unless it is supported by strong business processes. If the business is not conversant with the AI technology, then getting it across to customers becomes all the more difficult.
Yatra Online is moving from being a data-centric company to a data-driven one. How do you make use of the behavioral data to predict travel trends?
Using the principles of data governance and data lineage, Yatra has developed a centralized data warehouse. This system provides the needed impetus to make data-driven decisions. User behavior is captured by the searches they perform, the features they click on Yatra’s portals, and the bookings they make. We also try to understand the cross-booking behavior where the user may book more than one product type (hotel with flight, or cab with a flight). Once we have this behavioral data, we use the three pillars of data science to predict travel trends: data analytics, Machine Learning, and LLM-backed GenAI.
Instead of showing you a generic list of top-rated hotels, Yatra predicts the properties you are most likely to choose on the first page of search results, significantly reducing your “time-to-book.” As discussed earlier, we use HRNN and collaborative filtering to predict use behavior. We collect massive amounts of search data. This data tells us about destinations being searched. We can then see if these destinations are getting booked. Any changes in top destinations from one month to another is analyzed. Next, we figure out if the new destination can be served.
Search represents demand. By analyzing searches, we see, for example, if demand is building up in Tier-2 and Tier-3 markets. Accordingly, our inventory is adjusted. We also monitor bookings and study travel trends on an hour-by-hour basis to see if there is a drop or surge in the bookings. Actions such as price adjustments are made accordingly.
Yatra has the advantage of possessing data for both B2C and B2E (business-to-enterprise) transactions. We analyze the travel behavior of corporate and retail customers and identify the differences. Many times, we source external data to predict compressed planning windows where demand outstrips supply.
Can you share insights into Yatra Online’s cloud-native architecture that allows handling sudden surges in traffic during high-demand travel seasons?
Yatra Online employs a hybrid cloud approach for its cloud-native architecture. We have purchased servers, and we also rent servers on the Google Cloud Platform. Our critical applications that may witness surges in traffic are hosted on Google Cloud servers. Our DevOps team is able to make adjustments to CPUs and RAM to handle a heavier load.
As the popularity of the data warehouse grew, data of different types (variety – Big Data) needed to be loaded. This required more disk space, RAM, and CPUs. Basically, a server upgrade was needed. Since we are on the cloud with rented servers, we are able to achieve this with a few keystrokes.
How do you ensure data security and user privacy, especially with the integration of AI-driven chatbots in the user journey?
Yatra Online invests seriously to maintain its status as a compliant company. It is listed both in the US and India. For this reason, and because our large customers demand, we are SOC1, SOC2, and SOX compliant. From a data security and user privacy point of view, we are now investing heavily in being Digital Personal Data Protection (DPDP) compliant. We encrypt, obfuscate and anonymize Personal Identifiable Information (PII) in our application databases. We access LLMs only through APIs, thus ensuring that LLMs do not get trained on our data. A major concern with using external LLMs is transferring internal data to the LLM. We ensure that we never transfer any of the business or personal data through ChatGPT, Gemini or similar chatbots. Furthermore, PII data is not sent to the LLMs. As is the standard practice in data security, our data is saved on encrypted disks.
All our chatbots undergo multiple Vulnerability Assessment and Penetration Testing (VAPT) exercises. Checking for prompt injection, loading of malicious scripts, XSS etc. is part of this exercise. Since our bots are using a Google Cloud server, it adds another layer of security to the architecture.
In the next 5 years, how will GenAI change the way customers interact with travel portals?
Predicting the future is difficult, especially for an industry like Travel that is prone to disruption. But having been in the Travel space, we have developed a sense of where the travel industry is headed. We will move away from the 300+ travel websites era to one of agentic commerce.
Searches will be through conversational discovery. Instead of filters (4 stars, Pool etc.), one will provide complex, natural language prompts: “Find me a family-friendly spot in Goa for November that has a kids’ club, is near a quiet beach, and fits a Rs 40,000 total budget including flights.”
The portals will show the top 3 to 5 hotels based on your search and past booking behavior. If any of your reviews suggested noise issues, then AI will filter out hotels with such a feedback.
AI agents won’t just recommend; they will execute. You might say, “Book that Goa trip using my frequent flyer miles for the outbound and the best cash deal for the return.” The AI will navigate the checkout flows, apply your loyalty numbers, and handle the payment autonomously (assuming technology and government laws allow frictionless payments: yes in the US but not in India).
For business travel, AI will read your meeting invites, cross-reference your company’s travel policy, and present you with an itinerary that is already compliant.
There will be proactive disruption management, and AI will suggest alternatives instead of just notifying of the delay.
Portals may start bundling experiences (airport transfers, meals, etc.) into a single price instead of an à la carte mode of booking.
One may be able to switch between speaking and typing, and may even combine an image to get to the options being sought. DIYA has this functionality built in where you can get details about the landmark and plan an itinerary around it.
VR/AR may be introduced to give you a feel of the place before you book it.
A lot of this may happen. A lot will depend on how user behavior changes. Only time will tell how AIL will affect this change.
