- There are already many websites with lists and directories of lawyers (eg. https://singaporelegaladvice.com/find-a-lawyer/, https://www.bestlawyers.com/singapore) - however, these are completely useless if you don't know what area of practice you need!
- As 2 non-law students (Kok Rui and Yetong) who randomly signed up for this hackathon, we realised that without friends or brothers (eg. Law Student Kok Chee) with knowledge in the legal domain, it is incredibly difficult to find the type of legal service one might need - which means that these rankings and directories.
What it does
- Our web app has two components
- The first and main component allows a user to write out their situation in natural language (English) - no legalese required! Then, we using AI to figure out exactly what area of practice their situation is most relevant to!
- The second component is a platform that allows users to see lawyers in that area of practice recommended by external sites (prototype), as well as lawyers who signed up on the Law what Law? platform. These lawyers can list out their services and area of practice and be transparent with their fees - be it fixed or hourly. Users can then filter by fees if they wish.
How we built it
built with Vue.js.
Instead of a traditional Backend service, we wanted to challenge ourselves by building the entire "backend" based off AWS products (and not just whack it on an EC-2 instance and call it a day). We successfully accomplished this :D - our app is completely serverless and doesn't use any traditional backend frameworks.
We really went all in on this:
- AWS Cognito for Authentication
- AWS SageMaker for Deep Learning - all Model Training, Tuning, and Text Classification was done in SageMaker
- AWS AppSync for a GraphQL API
- AWS DynamoDB for our Database
- AWS Lambda to tie everything together (and make the site actually functional)
- First, we trained a BERT model on normalised LexisNexis text so it learns "legalese"
- Then, we scraped SAL's LawNet for all sorts of legal texts related to each area of practice we targetted. This gave us nicely-labelled data
- Classify!! (Transformers, TF-IDF, Sentiment Analysis, Word Associations)
Challenges we ran into
- One of our members fell sick halfway through the hackathon, so we were down to 1 programmer and 1 law content creator
Accomplishments that we're proud of
- We made something that actually works and is actually useful.. like.. right now.
- Learning new tech! We had pretty much 0 knowledge in every parts of the tech (Vue, AWS services, NLP) used in this project.
What's next for Law what Law?
- More training on more data!
- More granularity - perhaps not just area of practice but individual legal services
- Not everything is done - need to finish up fetching lawyers from external sites + making the profile editing page for lawyers that have joined our platform look a lot nicer.