Live legal case summariser

This product is our team's entry to HackTrinity 2024 GenAI x Legal, where it won the Most Viable Product (MVP) prize. We built a web application that provides instant summary of live cases, AI opinion on the cases and allows users to bet on the outcome of the case.

Project Schema

Technical Details

Setting up local instances:

You must have a PERPLEXITY_API_KEY defined in api/.env for this to work.

## installation
$ cd live-app-sum
$ npm install
$ cd ..
$ python3 -m venv .venv
$ pip install -r requirements.txt
$ source .venv/bin/activate

## running the services requires three terminal windows
### window 1
$ npm run start --prefix=live-app-sum

### window 2
$ cd bet
$ python betserver.py

### window 3
$ cd api
$ python app.py

Go to http://localhost:3000 and select a case.

The intention of this is to work with a live video / audio stream that is processed on the fly. For our proof of concept, we load the audio file of one of OJ Simpson's court appearances (api/oj1.wav). This file was 2 hours long so is processed in chunks of 1 minute. Each chunk is sent to a perplexity model (llama-3.1-sonar-large-128k-online) for a summary. Once a summary is completed, it is sent via a websocket notification to any clients subscribed to this particular court case. At the same time, the summaries are piped to the same model with a different prompt to provide an "AI expert opinion". These opinions are also published to the clients with websockets.