Neethi App 🏛️

AI-Powered Indian Legal Assistant for Common Citizens

Overview

Neethi App is a multi-agent AI system that provides accessible legal guidance to Indian citizens. Built using CrewAI for orchestration, InLegalBERT for legal text understanding, and Qdrant for semantic search.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        OFFLINE PIPELINE                         │
│  (Run once on GPU: Data Cleaning → Chunking → Embedding → Index)│
└──────────────────────────────┬──────────────────────────────────┘
                               │
                               ▼
┌─────────────────────────────────────────────────────────────────┐
│                      QDRANT VECTOR DB                           │
│                    (150K+ legal chunks)                         │
└──────────────────────────────┬──────────────────────────────────┘
                               │
                               ▼
┌─────────────────────────────────────────────────────────────────┐
│                        ONLINE PIPELINE                          │
│  Query → Librarian (Retrieve) → Lawyer (Reason) → Response      │
└─────────────────────────────────────────────────────────────────┘

Features

• Legal Advice: Get guidance on IPC, BNS, and relevant case precedents
• Document Drafting: Generate FIRs, RTI applications, and legal notices
• Location Services: Find nearby police stations, courts, and legal aid centers
• Legal News: Stay updated with latest judgments and law changes

Project Structure

neethi-app/
├── data/                      # Datasets (excluded from git)
│   ├── raw/                   # Original downloads
│   └── processed/             # Cleaned, chunked data
├── models/                    # Fine-tuned models (excluded from git)
├── agents/                    # CrewAI agent definitions
├── crews/                     # Multi-agent orchestration
├── pipelines/                 # Data processing scripts
│   ├── offline/               # One-time GPU tasks
│   └── online/                # Live query processing
├── api/                       # FastAPI gateway
├── configs/                   # YAML configurations
└── tests/                     # Unit tests

Quick Start

1. Environment Setup

# Create conda environment (Python 3.10 recommended)
conda create -n neethi python=3.10
conda activate neethi

# Install dependencies
pip install -r requirements.txt

# Download spaCy model
python -m spacy download en_core_web_sm

2. Configure Qdrant

Create configs/qdrant_config.yaml:

qdrant:
  url: "https://your-cluster.qdrant.io"
  api_key: "your-api-key"
  collection: "neethi-legal-kb"

3. Run Data Pipeline

# Step 1: Clean IPC/BNS data
python pipelines/offline/01_data_cleaning.py

# Step 2: Create chunks
python pipelines/offline/02_chunking.py

# Step 3: Populate vector database
python pipelines/offline/04_populate_qdrant.py

4. Start API Server

uvicorn api.main:app --reload --port 8000

Datasets

Dataset	Source	Size	Purpose
IPC Sections	Kaggle	500 sections	Statute lookup
BNS Sections	Kaggle	358 sections	New penal code
SC Judgments	Kaggle	26,000 PDFs	Case precedents
IndianBailJudgments	HuggingFace	1,200 cases	Bail prediction
IndicLegalQA	Mendeley	10,000 Q&A	Fine-tuning

Tech Stack

• LLM: Llama-3-8B (4-bit quantized)
• Embeddings: InLegalBERT (fine-tuned)
• Vector DB: Qdrant Cloud
• Orchestration: CrewAI
• API: FastAPI
• Deployment: Lightning AI (T4 GPU)

License

MIT License - Educational/Research Use

Team

Developed as part of [Your University] project.