Everyone likes to go travel during the summer and relax while they can. But how can you when you're leaving behind the important stuff that you couldn't fit in your 10 carry-on bags? Our team went through a similar thought process about this subject and we determined that there was a market segment in the insurance industry who probably had the same concerns as us. That's how the core of our idea was generated.
What it does
Travis is a smart chatbot specialized to insure your personal items, wherever you go. The goal of Travis is that you don’t have to call anyone or worry about anything. Before leaving for a trip or going on a vacation, you can have a quick conversation with travis about the things you'd like to insure and it'll give you an estimated price and a quote for your items.
How we built it
Tech-wise, we classify our objects with AWS Rekognition, and then we take that classification and scrape the web for the price of the object. We use AWS Lex for the chat-bot, however, we relay everything through a Flask backend so we can intercept the chat when we need to do specific things like upload images or calculate premiums. From the information, we gather from the user and automatically we determine the risk of insuring and item, and determine a premium to generate strong returns based on the specific policy. This is designed to be further augmented with machine learning once a dataset is generated to produce even more accurate returns. The front-end is in React and the backend is Flask.
Challenges we ran into
We ran into numerous problems during our time building travis. One of our major pain points sending and receiving image data from our React Frontend to our Python Backend. We eventually got the configurations for our call working and were able to send a string representation of the image to our Flask backend. Then the image was processed and a response was sent back with the information we extracted.
Accomplishments that we are proud of
Getting a project with this many moving parts working was a really big accomplishment for us. We were ambitious with our planning but we had faith in our abilities to execute them. Getting the three major components of our app interacting with each other meant that we could focus more on user experience and the dynamic behaviour of the chatbot.
One of the things we want to improve on is our machine learning model. We want it to estimate prices more accurately and it should be able to learn from the mistakes it makes. We also want our chatbot to cover a wider variety of user input and respond to it.