-
The training of our models, using Microsoft Azure, and our generated datasets.
-
The deployment of one of our Neural Networks. The web service will take a profile as an input and output a recommendation.
-
The SQL database, hosted on Microsoft Azure, which we can then access and modify from our application.
-
The queue of customers, as shown in the app
-
The view of pre-screening questions from the app.
Inspiration
We were inspired by the "closer to the customer" doctrine that T-Mobile has been pushing. We want to provide a way for representatives to provide the best possible experience to the customer, and one way of achieving that is to provide quality recommendations and improve convenience for the consumer. We were also inspired by Apple ,by focusing on streamlining the app, we increase the likelihood that represenatives will want to use the application and reduce the training needed by representatives to use the app.
What it does
Our application is an android app built for the reps, to better assist the consumers. The app allows representatives to queue up customers by having them scan their phone on an nfc reader located in store or alternatively manually queue up customers. The queue will be connected to an external database, hosted by Microsoft Azure. Next, when it's time for any particular customer, the representative should fill out the pre-screening fields for the customer. This will personalize the algorithm for the customer's specific needs. Afterwords the app will connect to Microsoft Azure and use the machine learning model that we trained in order to create personalized recommendations for plan, phone, features, and accessories. The rep can then use this information to better assist the customer.
How we built it
We built most of our application using a combination of Microsoft Azure and Android Studio. We trained our models in Microsoft Azure using data that we generated in Python. The models, along with our SQL database are then hosted in a web service on Microsoft Azure which can be accessed from our application. For the application we focused on the UI, making the process from queue to screening as seamless as possible.
Challenges we ran into
At first we ran into a hurdle getting enough data for our machine learning models. We later discovered that we can fix this problem by generating our own datasets using Python. We also ran into lots of problems reading the NFC input through our laptops because there wasn't a clear cut way on how to access the scanner and problems utilizing Microsoft Azure from Android Studio as there wasn't really a guide out there for Android, most of the sample code was in C++ and Python and didn't involve Android Studio.
Accomplishments that we're proud of
We're very proud of our machine learning models. We feel like they are extremely accurate and would allow for the best recommendations possible given a customer profile. We're also proud of the UI of our application because we really focused on making it look good and work seamlessly, which it does both.
What we learned
We definetely learned a lot about using Microsoft Azure and integrating that into different applications. We learned of ways to overcome lack of data and the process to designing a Machine Learning Algorithm. We were also suprised to learn all the potential solutions for customers that T-Mobile provides and which subsets of customers those solutions apply best to.
What's next for RepPrep
Next steps for RepPrep involve deploying this application to the app store for T-Mobile reps to start using and integrating the application with the NFC Scanner. Our application, at the moment, is mostly based in the cloud so it can be easily expanded nationwide with little to no restructuring of the app involved. We can expand our approach of using machine learning in customer interactions by integrating a person's Tmobile profile and past history to predict what they will need at any given time instead of only using it to recommend solutions.
Log in or sign up for Devpost to join the conversation.