Vichark LLM Model – Team LexCodex

Vichark is a multilingual, Retrieval-Augmented Generation (RAG)-based Large Language Model developed by Team LexCodex for the Smart India Hackathon.
It is trained and fine-tuned on The Constitution of India (English, Hindi & Odia) and related legal datasets, enabling citizens, lawyers, judges, and legal aid providers to access simplified, accurate, and multilingual legal information.

---

🚀 Problem Statements Covered

Our solution integrates and addresses five SIH problem statements:

1. SIH1700 – AI-based Interactive Chatbot for Department of Justice
2. SIH1701 – AI-Driven Research Engine for Commercial Courts
3. SIH1702 – Bail Reckoner Tool for Undertrial Prisoners
4. SIH1703 – Gamified Constitution Learning Platform (Citizen Perspective)
5. Multilingual Constitution Access – English, Hindi, and Odia

---

🛠️ Features

• Multilingual Support: English, Hindi, Odia
• Legal Q&A Chatbot – Answers user queries in a conversational format
• Commercial Court Research Engine – Case law aggregation & precedent analysis
• Bail Reckoner – Automated bail eligibility check
• Gamified Learning Module – Fun & interactive way to learn the Constitution
• Vector Database Search – Fast and context-aware retrieval
• Predictive Analytics – Insights based on historical legal trends

---

📂 Tech Stack

Backend:

• Python, FastAPI / Flask
• Hugging Face Transformers
• LangChain
• PyTorch
• FAISS Vector Database
• PostgreSQL / MySQL

Frontend:

• React.js / Next.js
• Tailwind CSS
• Streamlit (for prototyping)
• Framer Motion (animations)

Data Processing:

• Pandas, NumPy
• OCR: Tesseract
• Translation: IndicTrans, Hugging Face models
• NER: spaCy, Indic NLP

---

📜 Workflow

flowchart TD
    A[Data Collection] --> B[Preprocessing: Translation, OCR, NER]
    B --> C[Chunking & Embedding]
    C --> D[Vector DB: FAISS + SQL DB]
    D --> E[RAG-based LLM App]
    E --> F[User Interface: Web / Chat / Gamified Learning]

Loading

---

👨‍💻 Team LexCodex – Roles & Responsibilities

Backend

• Manas R. Das – Model Training
• Suryakanta Senapati – Data & Server Management

Frontend

• Sanskritee Behera – UI/UX Design
• Subhashree Prusty – Animations

Data Processing

• Suryakant Swain – Data Collection
• Dibyarani Behera – Data Preparation

---

⚙️ Installation & Setup

# Clone repository
git clone https://github.com/lexcodex/vichark-llm.git
cd vichark-llm

# Create virtual environment
python -m venv venv
source venv/bin/activate   # On Linux/Mac
venv\Scripts\activate      # On Windows

# Install dependencies
pip install -r requirements.txt

---

🚀 Running the Application

# Start backend server
uvicorn app:app --reload

# Start frontend
cd frontend
npm install
npm run dev

---

📌 Usage

• Select your preferred language (English/Hindi/Odia)
• Ask legal or constitutional queries in natural language
• Receive AI-generated, law-referenced answers
• Explore Bail Reckoner & Commercial Court Research Engine features
• Play Gamified Learning modules to learn the Constitution

---

📄 License

• This project is for educational and hackathon purposes under the Smart India Hackathon 2025.
• Legal accuracy is best-effort and should not replace professional legal advice.

---

🤝 Acknowledgements

• Department of Justice, Government of India
• Hugging Face
• Smart India Hackathon 2025 Organizers
• OpenAI & Indic NLP Resources