Andreas Kollegger is Director of Applied AI Research at Neo4j, developing graph-based approaches that help AI systems understand organizational context, assess risk, and make accountable decisions. His career spans NASA science missions, healthcare informatics in Zambia, and more than a decade building and explaining graph technology.
Kollegger encountered graph databases while developing medical informatics systems in Zambia, then joined Neo4j’s early core-engineering organization. He contributed to the Neo4j 1.3 manual and moved into product experience design by 2013. Subsequent responsibilities included developer advocacy and generative AI leadership before his transition into applied AI research. His early work on making relationships between records useful now informs his approach to retrieval, organizational memory, and autonomous decision-making.
- Build knowledge graphs around real questions. Start with generic entity labels, simple relationship types, and representative data; evaluate whether the system answers its intended questions before refining the schema or expanding coverage.
- Make relationships part of retrieval. Connecting document chunks exposes information unavailable from isolated passages, including relationships and graph communities. His approach to graph-enhanced retrieval distinguishes a vector index from a complete search system: graph properties, traversal, and post-filtering determine which results are appropriate.
- Give agents explicit decision boundaries. His framework for decision-aware AI agents considers objectives, prior decisions, organizational rules, risk, reversibility, and the cost of mistakes. Specialized agents can propose alternatives while separate agents assess their authority to act, escalate uncertain decisions, and preserve reasoning traces and decision precedents.
Kollegger’s agentic knowledge-graph workshop demonstrates multi-agent knowledge-graph construction. His aie-matrix project explores a virtual conference populated by autonomous personal agents. Both develop his central concern: giving AI systems enough connected knowledge, policy, memory, and authority to act responsibly.
