Building Agentic Knowledge Graph from Scratch

Transforming structured and unstructured data into a knowledge graph enhances LLM accuracy by linking data chunks to extracted entities and relationships. This multi-agent architecture utilizes Google’s Agent Development Kit (ADK) to automate the construction process, moving from user intent definition to schema proposal and final graph generation. Specialized agents act as intelligent control flow operators, delegating tasks such as file selection and entity extraction to ensure modular, scalable workflows. By integrating Neo4j for graph storage and implementing stateful memory management, the system maintains session context across asynchronous agent interactions. This approach replaces complex join-table queries with semantic pattern matching, enabling powerful access patterns for root cause analysis and recommendation engines. The framework provides a robust, adaptive environment for building complex, data-driven applications that bridge the gap between natural language and structured database logic.