Process-aware knowledge graphs boost reliability of LLM agent API calls, preprint says

[arXiv] A preprint by Saiful Khan and Afrah Farea says researchers integrated Open Data Protocol (OData) metadata and Business Process Model and Notation (BPMN) reference processes into a typed property graph to address a limitation in retrieval-augmented service discovery, which treats endpoints independently and misses process and schema constraints. A hybrid retriever identifies candidate endpoints using dense embeddings and follows process edges to retrieve subsequent endpoints. On 42 enterprise services with process-chain ground truth, the method improved Recall@3 by 5.7 points (n=74) for queries with low lexical overlap between the query and downstream entities, with consistent gains across embedding models and cut-offs; it matched dense retrieval on other queries and showed no improvement on OpenAPI services without a process layer, according to the paper. On 81 multi-step queries, an LLM agent completed 38.7% of tasks, versus 10.3% for a chunk-based baseline. In an elicitation probe, providing valid code lists eliminated invented codes but left invalid keys in 46% of write runs; full typed context reduced invalid calls to 3% (3 of 101 runs), with no invented codes. The authors say exposing process and schema constraints during retrieval improves the reliability of agent-driven API calls. The paper is a preprint and has not been peer-reviewed.