Latest News

Dr Ankur Narang Of Yatra Goes In Depth About The Technology Behind YUVA, Their Virtual Travel Assistant

Yatra, one of India’s leading online travel agencies and travel search engine launched its virtual assistant YUVA recently. This reinforces Yatra’s focus on the world of personalisation and delivering enhanced experience to its customers. Analytics India Magazine interacted with Ritesh Srivastava, who is the General manager at Yatra and Dr Ankur Narang, who heads Yatra Labs and leads Decision Sciences team, that brought into life the first-of-its-kind “hybrid approach-based bot” that allows user to use both voice and text to communicate easily, while simultaneously keeping the UI functionality.

YUVA is currently available on desktop, Android, iOS, Google Assistant and Facebook Messenger, but Dr Narang is ambitious about bringing it into other platforms like Alexa, Google Home, WhatsApp among many more in the near future. He says, “We are at a full throttle to open it up on all the platforms to maximise our reach”.

YUVA: The Idea And The Journey

The team started working on YUVA about six months ago, and ramped up their speed some three months ago. “With an initial focus on domestic and international flights, our main idea was to provide something different to our clients in terms of leveraging our current user interface, so that user is able to navigate through it easily and use voice and text for chat interactions,” said Dr Narang. “It gives a feeling to clients as if the virtual assistant is sitting in front of him/her and is able to answer all the relevant queries while making a booking,” he adds.

YUVA basically follows the scope of discussion and context to find an intent and accordingly review flight availability between source and destination. Srivastava says that as of now, YUVA can go from normal queries such as booking flight tickets from New Delhi to Mumbai, to longer queries like booking a non-stop flight for two between point A to point B via airline C in the morning within a certain price range. It filters the results one by one, and if a user is unable to speak due to noisy environment, he/she can type the query.

This is the version that is live now, and the team at Yatra wants to develop advanced versions where YUVA will be able to answer complex queries like details about all the past bookings, the amount of e-cash in the account, among multiple others. It has been worked upon by a small team of smart engineers who leverage open source technologies such as Python and databases like Cassandra and Aerospike, amongst others.

Making Of YUVA: Artificial Intelligence, ML, NLP

Dr Narang shares, “If we look at the functionalities mentioned above, the main task of YUVA is to understand the instances being communicated by the humans. For example, if there is a statement like, ‘Book a flight from Delhi to Mumbai via Indigo in the morning, departing tomorrow,’ there is a clear intent of booking a flight and that needs to be inferred. It has to accordingly filter the flights depending on departure and arrival cities, time specification, return flight, price range etc,”

“These entities are inferred by our slot filling algorithm where we use our in-house NLP module that identifies this intent and the details to be filled. As the conversation goes, it also has to maintain the context of the conversation. If you, for instance change your booking destination from Mumbai to Goa in this case, it has to identify it and perform a modified search. This is where AI comes into play that we have in our proprietary backend,” he said.

How it responds to every context, essentially needs to be learnt by the system and this is driven by Artificial Intelligence and ML.

Future Plans For YUVA

While YUVA is efficient in answering complex queries, there are further improvements that the Yatra team is trying to bring in, such as its ability to answer FAQs in a more interactive fashion rather than “just responding with some information” to the user. “We also want to expand it to other dimensions such as hotels, holidays, activities and others along with a strong touch of regional language coverage”, said Srivastava.

In fact, they also plan to do this by expanding from the current English language to ‘Hinglish’, Hindi, and major regional languages including Telugu, Tamil, Gujarati, Marathi and Punjabi, among others.

The Yatra team also wants to bring in more personalisation to give enhanced customer experience. “Most companies on travel have chatbot and virtual assistants but they are not on all platforms like us. We aim to be on every platform that is available in Indian market and have a broad umbrella, when YUVA is concerned”, says Dr Narang.

Personalising YUVA: Recommendation Engine Is The Key

Having stated the ambitious plan of bringing personalisation and understanding user behaviour, Yatra intends to have a full-blown recommendation algorithm which would use ML to drive various use-cases across flights, hotels and holidays. From churning campaigns to web personalisation, the recommendation engine would play a key role. “It would give us business preference, business guidance and motivate users to come back to the website — all of this in real time”, says Dr Narang.

Right now YUVA uses basic personalisation such as greeting a user as soon as he or she logs in; but they are working on a module where they could use a powerful model of the user behaviour and pro-actively give details and recommendations based on user choice and preferences. “We plan to have this feature coupled to our framework in a month’s time so that we could guide users to a much faster booking cycle based on their interest”, he shares.Also Read Interview – Nathan Falkenborg, Head of Analytics, APCEMEA at Visa

Other Areas Where Yatra Is Using AL, ML And Analytics

One of the key areas where Yatra is using AI and ML is dynamic discounting. This is to provide the best operational efficiencies and maximise the revenue with higher margins. “One needs to have these technologies in place as there are tens of thousands of hotels across the country. We need to be able to change their price based on multiple criteria such as length of stay, booking window, inventory availability, elasticity of city, user behaviour, amongst others.”

Dr Narang added that it is a complex problem and involves both prescriptive and predictive analytics — to be able to first accurately predict the performance of a hotel across multiple dimensions and then optimise the performance of the overall system dynamically to meet the overall business objectives. “This is done through an algorithm which involves flow orchestration of various types of data and performing multi-modal inference, leading to higher business impact for the company”, he said.

“The other area where we are using AI is essential is in the paid marketing sector, where we are tuning the bidding on platforms such as Adwords for multiple lines of businesses starting with flights, to improve performance across various dimensions,” he shares. “It involves a combination of business logic tuning, NLP, social network modelling, optimisation and reinforcement learning. This helps to dynamically perform ad tuning and auto-tune the bidding strategy to improve the performance of the ad,” said Dr Narang.

This is fundamental for e-commerce, as there is a typical dependence around discounting needs.

Challenges In The Space

Dr Narang shares that finding right talent is one of the key challenges. In addition, he shares that there are challenges such as data inconsistency and noisy data leading to large amount of time spent in data cleaning, among others. “It takes a lot of time to set a flow and crunch input data to bring it in the right format”, he says.

Dr Narang says that one has to be careful in the data cleaning aspect, as errors may lead the model into mis-inference, therefore reducing the accuracy and affecting the overall effectiveness of the output. He said that data cleaning necessity along with input data pattern changes require one to have a continuous check on the accuracy and performance of AI driven applications.

Focus Areas For The Company In 2018

Dr Narang is clear about the part where they continue work in areas like AI and ML, which are the key portfolios for the company. They plan to continue working on discounting, marketing spending to enhance efficiency and scale along with Virtual Assistant and Real-time context aware Recommendation Engines.

“We would also like bring best of technologies in AR/VR, deep sentiment analysis, blockchainand dive more into fraud management and cyber security aspects” shares Dr Narang.

On A Concluding Note

Dr Narang shares that it is important to deliver a strong experience to customers and drive higher customer retention and conversion and therefore investment in New Tech plays a vital role in giving a boost to the travel industry. It is the base of the pyramid of things like virtual assistants and its connect to various region and cultures could bring a valuable change in the hospitality industry in India.

“Yatra Labs is driving the company on a technology inflection point and Yatra will continue to deliver best in class products for its customers. We aim to have world-class ranking in terms of travel technology,” he says signing off.