| ### **Nepal Justice Weaver** | |
| --- | |
| Making the **three core features** (Law Explanation via RAG, Bias-Free Letter Generation, and Bias Auditing via EquityLens) **independent yet interconnected** creates a flexible, user-friendly platform. Users can access any feature alone or let the AI guide them through a seamless flow (e.g., explanation → detect need for letter → generate bias-free version). | |
| This modular approach strengthens the hackathon MVP: It's easier to build/test separately, but the connections make it feel intelligent and impactful. | |
| **Key Real-World Context (for Authentic Demo):** | |
| - Nepal's citizenship laws (Constitution Article 11, Citizenship Act 2006 with amendments) still have subtle gender biases: Easier via father's lineage; mothers often need extra declarations if father is "unidentified." | |
| - Common bureaucratic need: Formal applications/letters to District Administration Offices (DAO) for citizenship certificates. | |
| - Templates: Often include placeholders for "father's name," which can be biased—our tool can flag and rewrite to "parent's name" or neutral alternatives. | |
| #### Three Independent Modes + Smart Connections | |
| 1. **Mode 1: Bias Auditing Only** | |
| - User uploads an existing document (PDF/Word/Docx, e.g., a drafted letter or policy). | |
| - AI scans for biases (gender, socioeconomic, etc.—tuned for Nepal contexts like paternal assumptions). | |
| - Outputs: Highlighted report, bias score, explanations, and suggested rewrites. | |
| - **Interactive Twist**: "Bias detected! Do you want a bias-free updated version?" → If yes → Generate revised document (using generation module). | |
| 2. **Mode 2: Letter Generation Only** | |
| - User selects a template (e.g., "Citizenship Application," "Complaint to DAO") or describes need. | |
| - Inputs personal details (name, situation). | |
| - AI fills template → **Automatically runs bias audit** → Outputs a **guaranteed bias-free letter** (with fixes applied). | |
| - Download as Docx/PDF. | |
| 3. **Mode 3: Law Explanation Only (RAG Standalone)** | |
| - User describes situation in natural language (e.g., "I need citizenship for my child but no father" in English (for now)). | |
| - AI retrieves relevant laws (from our vector DB of Constitution, Citizenship Act, Muluki Civil Code snippets). | |
| - Explains in simple, plain language. | |
| - **No proactive linking here** unless needed. | |
| **Smart Connections (Conversational Flow):** | |
| - The app uses a **chat-like interface** to detect intent and offer cross-feature help. | |
| - Example Full Flow (Citizenship Scenario): | |
| 1. User: "I need citizenship but no father" → Mode 3: Explains law (e.g., "Per Nepal Constitution Article 11 and recent amendments, a child can get citizenship via mother's lineage with declarations if father unidentified. You may need to submit an application to DAO with birth certificate and affidavit."). | |
| 2. AI proactively: "This often requires a formal application letter to the District Administration Office. Would you like me to generate one for you?" | |
| 3. If yes → Switch to Mode 2: Ask for details (child's name, mother's details, district) → Generate letter using template. | |
| 4. Auto-run Mode 1: Audit for biases (e.g., flag any "father required" fields) → Apply fixes → "Here's your bias-free letter, rewritten to use neutral language." | |
| This makes it feel like a helpful AI assistant, not just separate tools! | |