Prototype of LegalEase Interface
Showcase
Actual working app

Inspiration

We felt that many times people don't have time to go through lengthy documents or the legal jargon is very tough to understand. That leads to agreeing on Terms & conditions which may feel unsatisfying later on. This issue inspired us to innovate a solution that focuses on legal documents only.

What it does

LegalEase is an innovative AI-powered tool designed to simplify the process of understanding complex legal documents, such as contracts, loans, and real estate agreements. It translates legal jargon into concise, easy-to-understand summaries by allowing users to upload images or PDFs of legal documents. Furthermore, LegalEase enhances user interaction through an integrated chatbot, enabling users to ask specific questions about the document and receive instant, understandable responses. This not only saves time but also empowers users with the clarity and confidence required for informed decision-making in critical legal matters.

How we built it

We used pre-existing LLM models and fine-tuned them on legal data. For training and tunning we used : OS (Operating System interactions) Torch (PyTorch library) Transformers (Hugging Face library for NLP) BERT (Bidirectional Encoder Representations from Transformers)

For GUI development Python, Figma, and Swift

Challenges we ran into

Fine tuning requires high computational power and takes too much time.

Accomplishments that we're proud of

We were able to generate meaningful and easy summaries of legal documents.

What's next for LegalEase

We are further going to simplify the process of uploading data on the platform. Will try to train the model in languages other than English, which will make it more inclusive. Generate both 'audio' and 'text' outputs.

Readme File:

LegalEase iOS Application

Team

Aman Ranjan
- Role: ML Model Training
- Contributions: Trained ML model using Large Language Models (LLMs), worked on the presentation (PPT).
Megha
- Role: ML Model Training
- Contributions: Trained ML model using LLMs, designed Figma app pages.
Prince
- Role: Backend Development
- Contributions: Implemented backend functionalities, worked on the presentation (PPT).
Shivam
- Role: Integration and App Development
- Contributions: Integrated the model and implemented the chatbot, developed Swift app generation.

Tech Used

Languages and Frameworks: Python, Swift
Design Tool: Figma
Libraries: Felt (UI/UX design), Torch (PyTorch for machine learning), Transformers (Hugging Face library for NLP), BERT (for NLP tasks)
Platforms: iOS (SwiftUI for UI, CoreData for local persistence)

Idea Description

How to Run the Project

Clone the Repository
- Clone the LegalEase project repository to your local machine.
Install Dependencies
- Navigate to the project directory in the terminal and install the required dependencies using CocoaPods or Swift Package Manager.
Open the Project
- Open the LegalEase.xcodeproj file in Xcode.
Run the Application
- Select a target device or simulator and run the application by pressing the play button or Cmd + R.

Core Code Explanation

iOS Applicarion

LegalEaseApp.swift: This is the entry point of the iOS application, where the main app structure and the persistence controller are defined. It sets up the initial view that the user sees upon launching the app.
Persistence.swift: Manages the local data persistence using CoreData, allowing the app to save and retrieve user data efficiently. This setup includes a container for CoreData entities and ensures that changes are merged from the parent context automatically.
ContentView.swift: The main interface of the app, presenting the upload button for PDF documents, displaying summaries, and facilitating user interaction with the chatbot. It leverages SwiftUI for a declarative UI approach.
NetworkManager.swift and GPTService.swift: Handle network requests to external APIs for generating summaries and chatbot responses. They encapsulate the logic for sending requests and parsing responses.

`model training using LLM'

***GPU Check and PyTorch Info:* Verify GPU availability; set device accordingly. Display CUDA status, device count, PyTorch, and CUDA versions.

File Reading and Dataset: Read 'judgement' and 'summary' files; store content in lists. Define LegalDataset using PyTorch for data tokenization and preparation.

T5 Model Fine-Tuning: Load pre-trained T5 model and tokenizer Define Adam optimizer, create dataset instances, and DataLoader. Fine-tune T5 on the legal dataset for three epochs. Save fine-tuned model in 'fine_tuned_legal_summarizer' directory.

Interruption and Considerations: Possible KeyboardInterrupt during training loop. Ensure ample resources for uninterrupted code execution. Consider runtime and system capabilities.

Model Utilization: Import necessary modules. Load original and fine-tuned models, and tokenizer. Save fine-tuned model and tokenizer. Specify text file path and read content. Tokenize and generate a summary using the fine-tuned model. Decode and print the generated summary.

Team Mates Details

Aman Ranjan: Specializes in machine learning and data science, with experience in NLP and model training.
Megha: A UI/UX designer and ML enthusiast, proficient in creating user-friendly app designs and working with NLP models.
Prince: A backend developer with expertise in API development, database management, and server-side logic.
Shivam: Specializes in machine learning and data science, with experience in NLP and model training. Focuses on app development and integration, skilled in Swift and SwiftUI for iOS development, and has a strong understanding of AI model integration.

Conclusion

LegalEase aims to bridge the gap between complex legal documentation and user understanding by leveraging AI technology to provide clear, concise summaries and interactive chatbot functionalities. Our team's diverse skills in machine learning, app development, backend development, and UI/UX design have contributed to creating a solution that addresses a common problem in legal document processing. We believe LegalEase will significantly impact how individuals interact with and understand legal documents, making legal advice more accessible and understandable.