SummaLegal
- #Legal
A solution for legal domain text document summarization using the OpenAI GPT-3/4 Davinci model. The goal was to create a set of scripts that could process legal documents of up to 25k words and generate a summarized output in a specific format.
- Natural Language Processing
Impact
Legal Document Summarization Solution:
- Algorithm to convert PDF into text and make short summarization with OpenAI GPT-3/4 API
Services we provided
Algorithm to convert PDF into text and make short summarization with OpenAI GPT-3/4 API
Tech Stack
GPT 3.5
GPT 4
Flask
PY2PDF
Challenges and Solutions
🧐 Challenges
- Text compression: Condensing the text by 50% while preserving critical information.
- Accuracy and quality control: Ensuring the accuracy of the compressed text through a rigorous quality control process.
💡 Solutions
We developed a solution that involved leveraging the power of the OpenAI GPT-3 Davinci model to perform a recursive chunk summarisation on the legal documents.
- The recursive chunk summarization algorithm broke down text into smaller chunks of sentences and summarised each chunk using the GPT-3 model
- Summarized chunks were combined and further summarised until the final output was achieved
- Output was formatted into three sections: “What this means,” “Why it matters,” and “Some other details that are relevant.”
User flow
1. User upload the PDF file (up to 25k words) with the information.
2. It is converted to text and proseccesed.
3. User gets a consise and accurate summary, organized into "What this means," "Why it matters," and "Some other details that are relevant" segments