⚖️ Nyay Mitra: Legal AI Assistant Platform

A comprehensive AI-powered legal assistant for Indian legal systems with intelligent document analysis, multi-language support, and advanced RAG capabilities.

---

Note: This GitHub repository is for demonstration and documentation purposes. The complete, production-ready components are deployed separately:

• AI Backend: Hosted on Hugging Face Spaces
• Node.js Backend: Handles authentication and database operations
• Frontend: User-facing application

✨ Project Overview

Nyay Mitra is a full-stack legal AI assistant specifically designed for Indian legal systems. The platform combines state-of-the-art AI technologies to provide document analysis, intelligent conversational assistance, multi-language translation, and automated document generation for legal professionals and individuals seeking legal information. Architecture: Frontend Application → Node.js Backend (Auth/DB) → AI Backend (Hugging Face) → External AI Services

🚀 Key Features

Intelligent AI Agent

• Multi-tool Reasoning Engine: Advanced workflow automation with LangChain orchestration
• Context-Aware Conversations: Session management with conversation continuity
• Automatic Language Detection: Smart routing for 10+ languages
• Multi-step Workflows: Complex query handling with intelligent tool selection

Document Processing

• RAG-Powered Q&A: Ask questions about uploaded documents with high accuracy
• Duplicate Detection: SHA-256 content hashing prevents redundant uploads
• Adaptive Analysis: Smart document analysis (small/medium/large strategies)
• Hybrid Retrieval: Semantic search with fallback mechanisms

Multi-Language Support

• 10+ Languages: English, Hindi, French, Urdu, Tamil, Bengali, Gujarati, Kannada, Malayalam, Telugu
• Bridge Translation: Direct and English-bridge translation paths
• Automatic Detection: Input language detection with confidence scoring
• Context-Aware Translation: Preserves legal terminology accuracy

Document Generation

• Template-Based Generation: Jinja2 templates for legal documents
• AI-Assisted Content: Automated summary and content generation
• Customizable Templates: Support for various legal document types

🛠️ Technology Stack

AI Backend (Hugging Face Spaces)

Component	Technology	Purpose
Framework	FastAPI 0.111.0	High-performance async API
LLM	Google Gemini 2.5	AI reasoning (Flash & Pro)
Orchestration	LangChain 0.1.20	AI workflow management
RAG Engine	LlamaIndex 0.10.34	Document indexing & retrieval
Vector Store	ChromaDB 0.4.24	Semantic search
Embeddings	Gemini text-embedding-004	Document vectorization
Translation	Argos Translate 1.9.0	Self-hosted translation
Storage	Hugging Face Datasets	Document persistence
Templates	Jinja2 3.1.4	Document generation

Full Stack

Layer	Technology
Frontend	NextJS
API Gateway	Node.js Backend
AI Processing	FastAPI on Hugging Face
Database	NeonDB
Authentication	JWT + OAuth 2.0

📋 System Architecture

┌─────────────────────────────────────────────────┐
│         Frontend Application (React/Vue)        │
│  • User Interface                               │
│  • Document Upload                              │
│  • Chat Interface                               │
└──────────────────┬──────────────────────────────┘
                   │
                   ▼
┌─────────────────────────────────────────────────┐
│         Node.js Backend (Auth/Database)         │
│  • User Authentication                          │
│  • Session Management                           │
│  • Database Operations                          │
│  • Request Routing                              │
└──────────────────┬──────────────────────────────┘
                   │
                   ▼
┌─────────────────────────────────────────────────┐
│      AI Backend (Hugging Face Spaces)           │
│  ┌────────────────────────────────────────────┐ │
│  │    Nyay Mitra AI Agent (Multi-tool)        │ │
│  └────────────────────────────────────────────┘ │
│  • RAG Service (LlamaIndex + ChromaDB)          │
│  • Translation Service (Argos Translate)        │
│  • Document Generation (Jinja2)                 │
│  • Deduplication Service (SHA-256)              │
└──────────────────┬──────────────────────────────┘
                   │
                   ▼
┌─────────────────────────────────────────────────┐
│        External AI Services & Storage           │
│  • Google Gemini API (LLM & Embeddings)         │
│  • Hugging Face Hub (Document Storage)          │
└─────────────────────────────────────────────────┘

⚙️ How It Works

Document Upload & Analysis Flow

1. Upload: A user uploads a document through the frontend.
2. Authentication & Routing: The Node.js backend authenticates the user and securely forwards the request to the AI backend.
3. Deduplication: The AI backend checks for duplicates using SHA-256 hashing to prevent redundant processing.
4. Processing Pipeline: If unique, the document is classified, chunked (800 characters with 150 overlap), embedded, and indexed in the ChromaDB vector store.
5. Response: An adaptive analysis is generated based on the document's size, and the results are returned to the user.

Conversational Query Flow

1. Query & Language Detection: A user sends a query, and the system automatically identifies the input language.
2. Workflow Planning: The AI Agent plans a multi-step workflow, selecting the appropriate tools (RAG, translation, etc.) for the task.
3. Reasoning & Execution: The agent executes the plan, manages conversational context, and generates a coherent response.
4. Display: The final response, translated if necessary, is displayed on the frontend with sources cited for transparency.

RAG Query Processing

1. Retrieval: The system performs a semantic search in the vector store to retrieve the top-k relevant chunks (max 12).
2. Context Formulation: A context is prepared for the LLM, managing a token limit of ~8,000.
3. Generation: The Gemini (Flash/Pro) model generates a precise answer based on the provided context and sources.

📊 Performance Specifications

• Chunk Size: 800 characters with 150 character overlap
• Max Chunks per Query: 12 (adaptive based on document size)
• Context Token Limit: 8,000 tokens (~32,000 characters)
• Analysis Token Limit: 20,000 tokens (~80,000 characters)
• Similarity Threshold: 0.3 for semantic search
• Deduplication: SHA-256 content hashing with in-memory cache
• Supported Document Types: PDF, TXT, DOCX

🌐 Live Deployments

AI Backend

• Live Demo: Hugging Face Space
• API Docs: Interactive Swagger UI
• Source Code: Hugging Face Repository

Frontend Application

• Live Application: Link to Frontend
• Repository: Link to Frontend Repo

Node.js Backend

• API Server: Link to Backend
• Repository: Link to Backend Repo

🔧 View Full API Endpoints

Primary Agent Interface

• POST /api/v1/agent/chat - Intelligent conversational interface
• POST /api/v1/agent/upload-and-chat - Upload & instant analysis
• GET /api/v1/agent/capabilities - System capabilities

Document Operations

• POST /api/v1/chat/rag - Document Q&A
• POST /api/v1/chat/rag/batch - Batch questions
• POST /document/suggest - AI analysis & suggestions

Translation

• POST /api/v1/translate - Text translation
• POST /api/v1/agent/detect-language - Language detection
• GET /api/v1/agent/languages - Supported languages

Deduplication

• POST /api/v1/agent/deduplication/check - Check for duplicates
• GET /api/v1/agent/deduplication/stats - Deduplication statistics

System

• GET /health - Health check
• GET / - System overview

Full API Documentation: Swagger UI

⚠️ Legal Disclaimer

This AI system is for informational and educational purposes only.

• ❌ Does NOT provide legal advice
• ❌ Does NOT replace qualified legal professionals
• ❌ Does NOT guarantee accuracy or completeness
• ✅ Provides general information and analysis
• ✅ Helps organize legal information
• ✅ Should be verified by legal experts

Always consult a qualified legal professional for specific legal matters.

📄 License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.

👨‍💻 Built By

Tejasvi Aryan && Anubrata Guin

📞 Contact & Support

For questions, issues, or collaboration:

• Email: tejasviaryan225@gmail.com

🙏 Acknowledgments

• Google Gemini for powerful LLM capabilities
• LangChain & LlamaIndex for AI orchestration
• Hugging Face for hosting infrastructure
• Argos Translate for open-source translation
• FastAPI for excellent async framework

⭐ Star this repository if you find it useful!