Each member of our team of 3 works part time on campus while taking a full credit course load. We have experience dealing with annoyed/frustrated customers all the time. We have also been on the other side, especially when you call up customer care, give your details, and hold for eternity while listening to jazz music. We live in the 21st century. There has to be a better way to do this. Now there is! That's how Aria was born.
What it does
Aria eliminates the constant back and forth communication need between customers and customer representatives/customer care agents. It solves the most common problems that users face, and once the root of the problem has been analysed, if needed it can escalate the situation to a human agent.
Aria also filters out mean/rude/derogatory comments/statements made by customers, so that customer service agents do not have to deal with that. By streamlining the process, Aria manages to keep both the customers as well as the agents happy.
How we built it
We used Wit.ai to assist us in building our Natural Language Processing library. We kind of built our own custom dataset, since Aria's utility is essentially across multiple industries and can dynamically change with API endpoints. For the backend, we used Node.js to run the server side application.
For the front-end, we decided to go with Facebook Messenger, since it is widely used and easily accessible. It also helps that most business have social media platforms and significant traffic flows through their pages. Their data such as comments on posts, etc. can also be used to solve problems/queries and streamline the customer service experience.
Challenges we ran into
We first used Dialogflow. Then we switched to Bitbot, since they had a challenge. Then after facing compatibility and permissions issues and realizing that we only had 6 hours to submit, we just decided to create our own backend and datasets and powered through the night.
Accomplishments that we're proud of
Finishing and submitting on time
What we learned
Don't use any framework without reading it's documentation inside out. Try to form a 4 person team because the one person can really make a difference when one of you'll gets tired and wants to sleep.
What's next for Aria
We plan to make this compatible with all industries, and deploy it on a public domain and a server so anyone can fork it. User data will only help the app, since it uses Natural Language Processing to become 'smarter' over time.