-
Mirror Image Logo
-
Web client
Mirror Image
About
Friendly chatbot that cheers users up by sending messages with positive sentiments. Asks thought-provoking questions that help users reevaluate their situation. Uses Microsoft Azure for sentiment analysis and language modelling.
Inspiration
We wanted to put together something that could help people who are feeling down or want someone to talk to.
What it does
Mirror Image helps users who are going through a rough day by being there for them. When the users talk about their days or feelings, the robot will prompt questions and comfort them.
How we built it
We built Mirror Image using Microsoft Azure, specifically their Language Understanding Intelligent Service (LUIS), Web Language Model API, and Text Analytics API. We fed the robot sample sentences and trained the robot with different types of responses. Also, we used flask and Heroku to deploy the robot online. Lastly, we created the Android version of this application.
Challenges we ran into
At first, we wanted to create a robot that takes in twitter feeds of a person and train to be like that person. This idea was inspired from the famous series "Black Mirror," Season 2 Episode 1, "Be Right Back." However, we faced some problems when we were trying to retrieve twitter feeds because the Tweepy API only retrieved to some finite number of tweets from the web. In addition, we could not train the robot to speak like the person because the robot kept on speaking gibberish after 40,000 iterations. Therefore, we decided to create the emotional support bot.
With the emotional support bot, we faced some problems trying to retrieve the data from the web onto the mobile devices. We first used Http and Unirest to do this; yet, it became very complicated to do POST requests. So we used OKHttp which was way simpler.
Accomplishments that we're proud of
The robot actually speaks and makes conversation! Not only that, the conversation makes sense. We added some jokes to the robot, so it is really nice to see when the robot tries to make terrible jokes. We are proud that we were able to deploy this robot to both web and android.
What we learned
We learned a lot about Microsoft Azure, their bot framework and their different bot training APIs. Also, we learned a lot about different types of responses a person can make from a single word.
What's next for Mirror Image
Our next steps include the following: - Continue working to implement functions such as allowing users with high sentiment to give inspirational quotations to those with low sentiment, Advising people with really low sentiment to visit free therapies.
Web client:
https://github.com/h-shan/mirror-image-web
Log in or sign up for Devpost to join the conversation.