README.md

metadata

title: Ese-ui
app_file: app.py
sdk: gradio
sdk_version: 5.32.1

Ese - AI Legal Assistant for Ghana

Ese is an AI-powered legal assistant specialized in Ghanaian laws, featuring voice input, text-to-speech responses, and comprehensive legal document analysis.

Features

• Voice Input: Web Speech API (Chrome/Edge) + Gradio audio fallback (Safari/Mac)
• Text-to-Speech: Edge TTS with African accent options
• Legal Knowledge Base: LightRAG-powered document retrieval
• Multiple Query Modes: Mix, Local, Global, Naive search
• Responsive UI: Clean Gradio interface with speech controls

Installation

1. Clone the repository:

git clone <repository-url>
cd ese-all-ui

2. Install dependencies:

pip install gradio lightrag-hku openai python-dotenv loguru markdown selectolax edge-tts soundfile numpy

3. Set up environment variables:

# Create .env file
OPENAI_API_KEY=your_openai_api_key_here

4. Prepare legal documents:

mkdir laws_storage
# Add your legal documents (.txt, .md, .json) to laws_storage/

Usage

1. Start the application:

python app.py

2. Open browser to http://localhost:7860

3. Voice Input: Click 🎤 Start Speaking (Chrome/Edge) or use audio recorder (Safari)

4. Text Input: Type questions directly

5. Audio Output: Select voice and enable speech output

Query Modes

• Mix (recommended): Combines local and global search
• Local: Specific entities and details
• Global: Broad concepts and themes
• Naive: Simple text matching

Voice Options

• British Female/Male
• US Female/Male
• Nigerian Female/Male
• South African Female/Male

Browser Compatibility

• Chrome/Edge: Full Web Speech API support
• Safari/Firefox: Audio file upload for transcription
• Mobile: Gradio audio recorder

Architecture

• Frontend: Gradio web interface
• RAG Engine: LightRAG with OpenAI embeddings
• LLM: GPT-4o-mini for legal analysis
• TTS: Microsoft Edge TTS
• STT: Web Speech API + file upload fallback

File Structure

├── app.py                 # Main Gradio application
├── laws_storage/          # Legal documents directory
├── legal_assistant.log    # Application logs
└── README.md

Configuration

Environment variables in .env:

OPENAI_API_KEY=sk-...           # Required: OpenAI API key

Legal Document Formats

Supported formats:

• .txt - Plain text
• .md - Markdown
• .json - Structured data

Documents are automatically processed with legal metadata for better retrieval.

Performance

• Response Time: 2-5 seconds
• Audio Generation: 1-3 seconds
• Knowledge Base: Scales to thousands of documents
• Concurrent Users: Supports multiple sessions

Development

For development mode:

python app.py
# Visit http://localhost:7860 with auto-reload

Add new documents and update:

python app.py --interactive

Limitations

• Requires internet for OpenAI API calls
• Web Speech API needs microphone permissions
• Safari has limited speech recognition support
• Audio files processed locally (no cloud upload)