Inspiration

One of the biggest problems during this COVID-19 pandemic and these awful times in general is that thousands of people are filing for property and casualty insurance. As a result, insurance companies are receiving an influx of insurance claims, causing longer processing times. These delays not only hurt the company, but also negatively impact the people who filed the claims, as the payout could be essential. We wanted to tackle these problems with our website, Smooth Claiminal. Our platform uses natural language algorithms to speed up the insurance claiming process. With the help and support from governments and businesses, our platform can save many lives during the current pandemic crisis, while easing the burdens on the employees working at insurance companies or banks.

What it does

Smooth Claiminal serves three main purposes:

  • Provides an analytics dashboard for insurance companies
  • Uses AI to extract insights from long insurance claims
  • Secures data from the claim using blockchain

The analytics dashboard provides insurance companies with information about the previously processed claims, as well as the overall company performance. The upload tab allows for a simplified claim submittal process, as they can be submitted digitally as a PDF or DOCX file.

Once the claim is submitted, our algorithm first scans the text for typos using the Bing Spell Check API by Microsoft Azure. Then, it intelligently summarizes the claim by creating a subset that only contains the most important and relevant information. The text is also passed through a natural language processing algorithm powered by Google Cloud. Our algorithm then parses and refines the information to extract insights such as names, dates, addresses, quotes, etc., and predict the type of insurance claim being processed (i.e. home, health, auto, dental).

Internally, the claim is also assigned a sentiment score, ranging from 0 (very unhappy) to 1 (very happy). The sentimental analysis is powered by GCP, and allows insurance companies to prioritize claims accordingly.

Finally, the claim submissions are stored in a blockchain database built with IPFS and OrbitDB. Our peer to peer network is fast, efficient, and maximizes data integrity through distribution. We also guarantee reliability, as it will remain functional even if the central server crashes.

How we built it

  • Website built with HTML, CSS, and JS for front end, with a Python and Flask back end
  • Blockchain database built with IPFS and OrbitDB
  • NLP algorithm built with Google Cloud's NL API, Microsoft Azure's Spell Check API, Gensim, and our own Python algorithms

Challenges we ran into

  • Setting up the front end was tough! We had lots of errors from misplaced files and missing dependencies, and resolving these took a lot more time than expected
  • Our original BigchainDB was too resource-intensive and didn't work on Windows, so we had to scrap the idea and switch to OrbitDB, which was completely new to all of us
  • Not being able to communicate face to face meant we had to rely on digital channels - this was exceptionally challenging when we had to work together to debug any issues

Accomplishments that we're proud of

  • Getting it to work! Most, if not all the technologies were new to us, so we're extremely proud and grateful to have a working NLP algorithm which accurately extracts insights and a working blockchain database. Oh yeah, and all in 36 hours!
  • Finishing everything on time! Building our hack and filming the video remotely were daunting tasks, but we were able to work efficiently through everybody's combined efforts

What we learned

  • For some of us, it was our first time using Python as a back end language, so we learned a lot about how it can be used to handle API requests and leverage AI tools
  • We explored a new APIs, frameworks, and technologies (like GCP, Azure, and OrbitDB)

What's next for Smooth Claiminal

  • We'd love to expand the number of classifiers for insurance claims, and perhaps increase the accuracy by training a new model with more data
  • We also hope to improve the accuracy of the claim summarization and insights extraction
  • Adding OCR so we can extract text from images of claims as well
  • Expanding this application to more than just insurance claims! We see a diverse use case for Smooth Claiminal, especially for any industry where long applications are still the norm! We're also hoping to build a consumer version of this application, which could help to simplify long documents like terms and conditions, or privacy policies.
Share this project:

Updates