Customer Representative Interface
Nobody likes holds. Representatives don't like leaving customers hanging, and customers don't like waiting for responses. We believe that customers should get answers right after they ask a question for the best experience. So we decided to make a real-time question finder for customer representatives that would search for solutions as the customer describes the problem.
What it does
Our system gives customer representatives the most relevant answers right as the customer finishes describing the problem. Rather than forcing the customers to hear a number scheme and having them describe their problem after a long queue, Liberate allows them to discuss their concerns immediately and have the representative aware of the solution upon a match.
How we built it
We used state of the art machine learning as our basis of keyword extraction, natural language processing and speech to text recognition. The server back-end architecture is built with node.js and Twilio, allowing customers to reach representatives with just a phone call and forming a well-structured pipeline thanks to multiple middle-ware servers. The front end consists of Vue.js to give the representative a page with all of the steps to the solution of the problem which the customer is facing.
Challenges we ran into
Getting speech to text to work for only one side of the call while allowing the representative and customer to communicate fluidly was among the greatest difficulties faced in the creation of Liberate. Additionally, to train our machine learning systems we needed data. Since past customer service interaction data was not available, we used the T-Mobile Support Discussion Board as our source. Scraping the information off of the T-Mobile site took time both in establishment and execution. The last obstacle we faced in development was in implementing a successful client-service system using socket.io in Node.js. We are also struggling greatly with the usage of a jar file. This was the last obstacle standing between us and success.
Accomplishments that we're proud of
We are proud of our machine learning system, natural language processing system, and automated real-time call question lookup system.
What we learned
We learned of the great difficulties that come with limited data. In order for Liberate to be complete, another application that automatically creates documentation for newly solved problems is necessary for Liberate to adapt.
What's next for Liberate
Thanks to Liberate's ability to fine-tune customer descriptions, we can diagnose problems much more efficiently