This document discusses using graph technology and natural language processing to harmonize cancer research data from different sources. It describes using GPT models to generate synonyms and parse text, representing the data as nodes and edges in a Neo4j graph, and calculating text similarity to link related concepts. This approach allows mapping between non-standard terms, correcting typos, and classifying nodes. Queries are run on the graph to identify related headers. An interactive GPT interface is proposed for graph management.