Archaeology often begins with clues hidden in terrain: a raised mound here, a linear depression there — the silent traces of a vanished world. This talk introduces Archaios, a graph-driven platform that uses Neo4j to represent, reason about, and orchestrate the discovery of archaeological features from LiDAR and historical data.

The system combines a semantic graph model of terrain features, place-names, and historic references with a multi-agent framework powered by Autogen. Agents coordinate tasks like terrain segmentation, image analysis via GPT-4 Vision, and classification of possible man-made features. When candidates are found, they’re logged in Neo4j as hypotheses and routed to archaeologists via a notification system for final confirmation.

You’ll learn how Neo4j serves as a central semantic hub — encoding spatial relationships, human feedback, and interpretive hypotheses — while agentic LLMs act as flexible analysts. We’ll explore how to combine structured and unstructured reasoning in a graph-native architecture, and how feedback loops improve both discovery and trust in AI-supported domains.