module_c/README.md

Module C: RAG-Based Letter Generation

This module generates professional, bias-free Nepali government letters by combining Retrieval-Augmented Generation (RAG) with LLM-based adaptation.

Instead of manually selecting templates, users simply describe their need (e.g., "I need a citizenship certificate for my child"), and the system:

1. Retrieves the most relevant official template from the vector database.
2. Generates a final letter by intelligently filling the template with user details.

Features

• Intent-Based Retrieval: Finds the right template even if the user doesn't know the official name.
• Smart Filling: Extracts details (Name, Date, District) from natural language.
• Modular Design: Separate Indexer, Retriever, and Generator components.

Setup

1. Install Dependencies:

   pip install chromadb sentence-transformers mistralai python-dotenv
   

2. Environment Variables: Create a .env file in the project root:

   MISTRAL_API_KEY=your_api_key_here
   

Usage

1. Python API

from module_c.interface import LetterGenerationAPI

api = LetterGenerationAPI()

# Describe your need in English or Nepali
description = "I need to apply for citizenship for my daughter Sita. I am Ram Sharma from Kathmandu."

result = api.generate_smart_letter(description)

if result['success']:
    print(f"Used Template: {result['template_used']}")
    print(result['letter'])
else:
    print(f"Error: {result['error']}")

2. Interactive Flow (Handling Missing Info)

# 1. Analyze Requirements
analysis = api.analyze_requirements("I need a citizenship certificate")

if analysis['success'] and analysis['missing_fields']:
    print(f"Selected Template: {analysis['template_name']}")
    print(f"Missing Fields: {analysis['missing_fields']}")
    
    # 2. Provide Missing Info
    additional_data = {
        "Date": "2081-01-01",
        "District": "Kathmandu",
        "Applicant Name": "Ram Sharma"
    }
    
    result = api.generate_smart_letter(
        "I need a citizenship certificate", 
        additional_data=additional_data
    )
    print(result['letter'])

3. Adding New Templates

1. Add your template text file (.txt) to data/module-C/.
2. Run the indexer to update the vector database:

   python module_c/indexer.py
   

Project Structure

• interface.py: Main entry point.
• retriever.py: Finds relevant templates using Vector DB.
• generator.py: Uses LLM to fill retrieved templates.
• indexer.py: Ingests templates into ChromaDB.
• vector_db.py: Manages ChromaDB connection.
• config.py: Configuration settings.