Jeanlebaptist/final-ai-project-2025

My final project (ClimateWatchGH ChatBot) was on nature conservation. The illegal small-scale gold mining (locally known galamsey) activities in Ghana are a threat to forest reserves, rivers, farmlands and human beings. As part of livelihood practice, it is becoming difficult to fight galamsey. This anti-galamsey ChatBot will help combact it. ClimateWatchGH: A Multimodal AI Agent to Combat Illegal Mining Overview ClimateWatchGH is an AI-powered, multimodal chatbot agent designed to assist in the fight against illegal small-scale mining, locally known as galamsey, in Ghana. The project addresses the critical environmental, social, and economic damage caused by these activities by creating an intelligent system that can process diverse data types—including text and images—to provide informed and timely responses. The agent's architecture is built on a robust pipeline that integrates data collection, model training, and a conversational interface, making it a powerful tool for researchers, policymakers, and environmental agencies. Key Features Multimodal Capability: Processes and responds to queries based on both text documents and images. Conversational Interface: Uses a Gradio-based chatbot for intuitive and easy interaction. Retrieve-Augment-Generate (RAG): Enhances the chatbot's knowledge by retrieving information from a curated corpus of documents and augmenting its responses with cited sources. Computer Vision: Employs a YOLOv8n model to detect and identify signs of illegal mining in images. Modular Pipeline: The entire project is structured into distinct, logical notebooks for data handling, model training, and application development. Project Architecture The project's architecture is orchestrated by LangChain, which acts as the central framework. It connects various components to form the complete agent: Core Reasoning: Leverages the GPT-4o mini model for natural language understanding and generation. Embeddings: Uses OpenAI Embeddings to convert text into numerical vectors for efficient similarity search. Vector Database: Pinecone serves as the vector store for the text-based RAG system. Object Detection: A fine-tuned YOLOv8n model is used for image analysis. User Interface: Gradio provides a simple yet powerful web-based UI for the chatbot. Data and Sources The project's effectiv …