During our deep-dive into problems with common-day customer service, we learned the importance of transcripts and their effect on customer understanding. A big problem within modern-day call centers and store locations is the fact that there exists a loss of information when transferring customers between mediums (customer-service agent --> supervisor, online --> in-store, etc.). This is usually solved with a transcript of a chat log/conversation but no one wants to trudge through an entire multi-page transcript to retrieve a handful of key-points.

What it does

Utilizing a plethora of azure tools, we are able to have a live transcription service during any call between a customer and service-representative. In this live script, we analyze the semantics (customer attitude) live so the agent know if he or she is performing well. In addition, we are able to extract key phrases through ML and use a summarization agent to highlight key-points within the transcript itself. By the end, we are able to not only compress a multi page transcript into around a dozen key-points for easy transfer between teams but also to provide the agent a live platform to see the customer's response to better adapt to the situation at hand. In addition, we can also now keep track of users average sentiment with not only their phone but T Mobile as a whole. Now, when entering a retail store, a representative can get a summary of the customer's most recent smilometer value/sentiment towards the company and prepare themselves mentally to aid in whatever way deemed worthy.

How we built it

We really took advantage of the Azure suite. Azure's speech-to-text API was efficient in conveying user voice input in punctuated text that our algorithms like semantic analysis could use. Similarly, the token algorithm uses azure keyphrase extraction and language detection in order to determine the sentences within the log scoring the highest. These scores are calculated based off of keyphrases, sentence length, number abundance and a lot of other factors fine tuned to highlight the best of sentences.

Challenges we ran into

We were very tight on time and were extremely ambitious. We practically have been programming our solution down to the wire due to the plethora of moving parts. Usually we try to get 2-3 hours of sleep during a hackathon but today we had absolutely 0.

Accomplishments that we're proud of

We were very proud of ourselves and our ability to efficiently work with so many novel technologies and successfully tie them together in the end. In the end, our proudest accomplishments fall into two features: the T-gist and the Smilometer. The T-gist attends to the need of universal access of transcripts/gists in order to help the customer. No one wants to repeat the same problems to multiple different people. Similarly, the Smilometer helps maintain the most important thing when it comes to t-mobile customers: relationships. With live semantic analysis, service agents can now look at their customer's attitudes real time and adjust their approach to aid.

What we learned

Azure was completely new to us; however, we wanted to dive in and use as much as we possibly could! We discovered that daisy chaining APIs is a true art form and that tying all of our moving parts together was much more strenuous and time consuming than anyone of us could have expected. However, saying that we are proud of our accomplishments and what we learned is an understatement!

What's next for T-Gist

We might move forward with T-Gist and modify the core gist functionality to apply to a plethora of different fields and companies. Only time can tell where we go with it!

Built With

Share this project: