Matthew Scardino

This project presents the development of a modular AI-enhanced system to process, classify, and retrieve wildlife images and videos. Integrates traditional computer vision techniques with advanced semantic understanding powered by AI models. The platform supports manual and AI-assisted annotation, stores visual metadata and embeddings, and enables intuitive natural language queries to discover relevant visual content. By enabling contextual insights and advanced search capabilities, the system transforms how wildlife media can be explored and utilized.

Version 1 of this project introduced a system that combined computer vision, metadata extraction, and semantic embeddings to organize and retrieve wildlife imagery. Version 2 builds on that foundation by expanding the platform into a comprehensive AI driven framework capable of deeper ecological reasoning and more accurate species identification. This iteration incorporates transfer learning, multimodal CLIP embeddings, supervised classifiers such as SpeciesNet, and YOLO object detection, integrated with PostGIS geospatial analysis and ecoregion aware ranking methods. The system also introduces LangGraph agent orchestration and a dedicated Wildlife MCP API, enabling dynamic tool execution, similarity reranking, and context sensitive inference. By unifying visual, semantic, and spatial signals within a modular architecture, Version 2 enhances classification accuracy, supports natural language exploration, and enables more meaningful ecological interpretation of wildlife image datasets.