Khadamet: AI-Powered Legal & Administrative Assistant for Tunisia

🏆 Project Overview

Track: Tunisia's Legal and Administrative Systems

Khadamet is an innovative AI-powered web application designed to democratize access to legal and administrative guidance in Tunisia. By bridging the gap between complex bureaucratic procedures and everyday citizens, Khadamet empowers users to navigate the transition from informal to formal work with confidence and ease.

📋 Project Summary

Khadamet serves as a comprehensive digital assistant that simplifies Tunisia's complex legal and administrative landscape. The platform offers:

Multilingual Support: Arabic, Tunisian Dialect, French, and English
Voice-Driven Interface: Speech-to-text and text-to-speech capabilities
Multimodal Interaction: Text, image, and video processing
Specialized Domains: Auto-Entrepreneur registration, CNSS enrollment, banking procedures, and legal consultations
Advanced AI Features: Document extraction, Retrieval-Augmented Generation (RAG), and specialized legal expertise

The application stores uploaded images in FAISS vectorstores, enabling context-aware responses through RAG technology. A distinctive "Legal Expert" feature connects users with a fine-tuned LLM specialized in Tunisian legal codes and laws.

Impact: Khadamet enhances accessibility for informal workers, reduces administrative burden on public institutions, and promotes inclusive economic growth while supporting Tunisia's digital transformation agenda.

🎯 Project Links & Resources

📺 Demo & Presentation

Demo Video (2-3 min): Watch on SharePoint
Presentation Slides: View on Canva

📚 Documentation & Code

Technical Report: Google Docs
GitHub Repository: View Code
Dataset: Download from SharePoint

🌐 Live Application

Live Demo: https://ains.khadamet.com

🎤 Elevator Pitch

"Khadamet is an AI-powered web platform designed to simplify the transition from informal to formal work for Tunisian entrepreneurs. By offering multilingual, voice-enabled, and multimodal support, it breaks down the complex legal and administrative procedures required to access the Auto-Entrepreneur system and essential services like CNSS and banking.

With features like document extraction, Retrieval-Augmented Generation (RAG), and direct access to legal experts, our solution empowers users—especially freelancers, street vendors, and delivery workers—with clear, accessible, and affordable guidance.

Khadamet not only reduces the burden on public institutions but also contributes to Tunisia's digital transformation and economic inclusion by making legal and administrative help available to all."

💡 Project Deep Dive

🎯 Inspiration

Our motivation stems from real-world challenges faced by Tunisians navigating the Auto-Entrepreneur system. Despite digitization efforts, the transition from informal to formal economy remains complex and discouraging:

Low Adoption Rates: Complex procedures deter informal workers from registering
Information Fragmentation: Follow-up steps (CNSS, banking, micro-loans) are poorly documented
Language Barriers: Information often unavailable in accessible formats
Cost Barriers: Legal consultations are expensive and inaccessible

We recognized Tunisia's multilingual context and digital transformation goals as opportunities to create an inclusive, AI-powered solution that democratizes legal guidance.

🎯 What Khadamet Does

Khadamet transforms how Tunisians interact with legal and administrative systems by providing:

Multilingual Guidance: Support in Arabic, Tunisian Dialect, French, and English
Voice-First Interface: Speech-to-text and text-to-speech for accessibility
Multimodal Processing: Text, image, and video input handling
Context-Aware Responses: RAG system for accurate, relevant information
Expert Consultations: Access to specialized legal LLM for complex queries
Step-by-Step Guidance: Visual and audio instructions for procedures

🛠 Technical Architecture

Backend Infrastructure

Framework: Flask
Embeddings: SentenceTransformers for text vectorization
Vector Search: FAISS for efficient similarity search
Architecture: Hybrid RAG (Retrieval-Augmented Generation)

Frontend Experience

Technologies: HTML5, CSS3, JavaScript
UI Libraries: Font Awesome 6.4.0, Google Fonts (Montserrat)
Markdown Rendering: Marked.js for rich content display

AI & Voice Integration

Voice Processing: Web Speech API with multi-language support
Multimodal AI: Groq API for image content extraction
Conversational AI: OpenAI API for natural language interactions
Language Detection: franc-min for automatic language identification

Specialized Features

Legal Expert Module: Dedicated consultation service (Port 5001)
Visual Guidance: Preloaded videos and images for complex procedures
Document Processing: Base64 encoding for secure file handling

🚧 Technical Challenges Overcome

1. Multilingual Voice Recognition

Challenge: Inconsistent browser support across Arabic, French, and English
Solution: Dynamic language detection with intelligent fallback mechanisms

2. Multimodal Performance Optimization

Challenge: Balancing response speed with processing quality across input types
Solution: Coordinated API calls with optimized caching strategies

3. Scalable Vector Search

Challenge: Managing FAISS indexes across diverse legal domains
Solution: Optimized chunking strategies and hierarchical vectorstore architecture

4. Accessibility Design

Challenge: Ensuring usability for non-technical users with varying digital literacy
Solution: Iterative user testing and inclusive UI/UX design principles

🏆 Key Accomplishments

Social Impact: Created a genuinely inclusive tool addressing real societal needs
Technical Innovation: Successfully integrated multilingual speech recognition with legal RAG
Specialized AI: Developed fine-tuned LLM for Tunisian legal consultations
User Empowerment: Built a platform that truly serves underserved communities
Accessibility: Designed for users with varying technical backgrounds and languages

📚 Learning Outcomes

Our development journey enhanced our expertise in:

Advanced RAG Systems: Deep integration of FAISS with SentenceTransformers
Multilingual NLP: Handling diverse language inputs and outputs
Multimodal AI: Coordinating text, voice, and visual processing
Accessible Design: Creating inclusive interfaces for diverse user bases
Full-Stack Integration: Synchronizing complex backend logic with intuitive frontends
User-Centered Development: Building solutions that address real-world problems

🚀 Future Roadmap

Short-Term Goals (3-6 months)

Knowledge Base Expansion: Integrate additional Tunisian legal documents
Model Enhancement: Further fine-tune Legal Expert LLM with specialized datasets
Partnership Development: Collaborate with public institutions and NGOs

Medium-Term Goals (6-12 months)

Mobile Application: Native iOS/Android apps for enhanced accessibility
Communication Channels: WhatsApp and SMS integration for broader reach
Offline Capabilities: Core functionality without internet dependency

Long-Term Vision (1-2 years)

Regional Expansion: Adapt platform for other North African countries
AI Advancement: Integrate latest language models and multimodal capabilities
Ecosystem Integration: Connect with government digital services and databases