Versa – AI-Powered Legal Document Transformation

Versa is a platform designed to make complex legal documents more accessible without losing legal precision. Unlike traditional summarization tools that either oversimplify or remain inaccessible, Versa focuses on context-preserving transformation — adapting legal text into different levels of readability while maintaining accuracy.

Problem Statement

Legal documents are inherently dense and difficult to interpret, creating barriers for individuals and businesses. Existing solutions either dilute critical legal nuance or remain unusable for non-experts.

Engineering Solution

Versa implements an AI-powered, multi-provider translation pipeline that allows users to transform legal documents into clear, accessible formats. The platform ensures:

• Accuracy Preservation: Legal meaning is retained across all transformations.
• Provider Abstraction: Multiple AI providers (OpenAI, Anthropic, Google AI) are supported through a unified abstraction layer.
• Scalable Frontend-First Architecture: Client-side parsing, secure backend orchestration, and real-time user feedback.

Core Engineering Highlights

• Multi-Provider AI Abstraction: Vendor-independent design for cost optimization, reliability, and feature flexibility.
• Dynamic Prompt System: Modular templates for different transformation contexts (plain-language, child-friendly, structured JSON).
• Real-Time PDF Pipeline: Hybrid client-server processing optimized for large documents.
• Security-First Key Management: AES-256 encryption, rate limiting, and session-based authorization.
• User-Centric Frontend: component-driven architecture with smooth real-time feedback and error handling.

---

Features

• React + Next.js frontend with client-side PDF parsing (PDF.js).
• API key management with AES-256 encryption.
• Integration with multiple AI providers (OpenAI, Anthropic, Google AI).
• Abstraction and fallback handling across AI providers.
• Error handling, retry logic, and request validation on the backend.
• Usage tracking and logging for user activity.

---

System Architecture

The system is structured into four layers: Client, API, Processing Core, and Persistence. This separation provides clarity of responsibility and supports future scalability.

---

Technology Stack

Frontend

• Frameworks: Next.js 15, React 19, TypeScript 5
• UI & Styling: Tailwind CSS, PostCSS, Autoprefixer, Framer Motion, heroicons
• File Handling: react-pdftotext (PDF parsing), react-dropzone (file uploads)

Backend

• Runtime & API: Node.js, Next.js API Routes
• Authentication & Security: NextAuth.js, Prisma adapter, bcrypt, AES-256 encrypted API keys
• Database: PostgreSQL with Prisma ORM

AI Integration

• Providers: OpenAI, Anthropic, Google AI, Mistral AI, Cohere
• Models: GPT-4o, GPT-4o Mini, GPT-3.5 Turbo, Claude 3.5 Sonnet, Claude 3 Haiku, Gemini 1.5 Pro, Gemini 1.5 Flash, Mistral Large, Command R+

Architecture Highlights

• Component-driven frontend with reusable UI modules
• Serverless API routes with input validation and rate limiting
• Client-side PDF parsing for privacy and reduced server load
• Multi-provider AI abstraction for vendor independence and reliability
• Secure session-based authentication with JWT and encrypted key storage

---

Getting Started

# Clone repository
git clone https://github.com/aviroopjana/versa.git

# Install dependencies
npm install

# Run locally
npm run dev