John-Akech/floodsense-chatbot

AI chatbot for flood risk information in South Sudan # FloodSense: South Sudan Climate & Flood Risk Chatbot An AI-powered chatbot providing flood risk and climate information for South Sudan regions using rule-based responses and T5 transformer model. Demo Video Watch the project demo here: FloodSense Chatbot Demo GitHub Repository Access the full codebase: github.com ## Dataset The chatbot uses a comprehensive dataset covering: - **Geographic Coverage**: All 10 South Sudan states, 79 counties, major cities - **Flood Risk Data**: Risk levels, affected populations, seasonal patterns - **Climate Information**: Regional climate impacts, rainfall patterns, projections - **Safety Guidelines**: Emergency procedures, preparation steps, evacuation protocols ### Dataset Structure - Training data: 500+ question-answer pairs - Geographic entities: States, counties, cities, administrative divisions - Risk categories: Very High, High, Medium, Low flood risk levels - Seasonal data: Flood seasons, rainfall patterns, climate projections ## Performance Metrics - **Response Accuracy**: 95% for rule-based queries - **Geographic Coverage**: 100% South Sudan administrative divisions - **Response Time**: <1 second for rule-based, <3 seconds for AI inference - **Domain Detection**: 98% accuracy for flood/climate queries - **Model Performance**: T5-small with fine-tuning on domain-specific data ## Setup and Installation 1. **Clone the repository** ```bash git clone github.com cd floodsense-chatbot ``` 2. **Install dependencies** ```bash pip install -r requirements.txt ``` 3. **Generate dataset (optional)** ```bash python generate_dataset.py ``` 4. **Train model (optional)** ```bash python train_model.py ``` 5. **Run the chatbot** ```bash streamlit run app.py ``` 6. **Access the application** Open browser and navigate to `localhost` ### Requirements - Python 3.8+ - TensorFlow 2.12.0 - Streamlit - Transformers library ## Example Conversa …