Inspiration
Bar exam takers often need to reference long records of cases to understand a particular law. To ease this process, a handy chatbot can help them quickly reference a particular law by just taking the natural language input instead of a specific law number which is hard to remember.
What it does
The application answers questions related to US laws and aids novice lawyers who are preparing for the bar exam. It also guides the user in case the question does not contain specific details that are required for an answer.
How I built it
We created a REST API that receives questions and extracts relevant question entities like the question type (eg: law for employee wages), sub-question type (eg: law for temporary employee wages), jurisdiction (eg: Alabama, New York) and filters the answers based on these criteria. If no match is found, then the closest answer is returned by calculating a similarity score using Tf-idf vectors. This API was consumed by a Django web app and an Android chatbot.
Challenges I ran into
We had originally intended to expose the API as an Alexa skill, however, integrating Azure LUIS API with the AWS Alexa skill turned out to be very difficult. We then decided to build a generic framework that can be called from any end-user device.
Accomplishments that I'm proud of
We were able to successfully build a product from scratch after our initial prototype turned out to be infeasible due to the time constraints.
What I learned
Modifying and massaging the data to be able to suit your model can prove to be more important than the model accuracy itself. Moreover, your product should be able to handle faulty and noisy inputs by the user.
What's next?
Building an Alexa skill for this application and making the answering system more robust.
Built With
- android
- azure-luis
- django
- javascript
- nltk
- python
Log in or sign up for Devpost to join the conversation.