FinOps over Heterogeneous Systems: Enabling Agentic Workflows via GraphQL

FinOps (Finance + Operations) involves analyzing and optimizing cloud infrastructure costs using data that is distributed across multiple heterogeneous systems, including cloud providers, observability platforms, and financial tools. These sources differ in schema, access patterns, and semantics, making it challenging to perform unified analysis and derive actionable insights.

In this paper, we present a FinOps agent designed to operate over such heterogeneous data environments. Our approach integrates a GraphQL-based abstraction layer to unify access across data sources, enabling structured and consistent data retrieval. On top of this unified interface, we leverage large language models to translate natural language queries into executable GraphQL queries and to orchestrate multi-step analysis workflows.

We demonstrate how this combination allows users to perform complex cost analysis and optimization tasks through natural language interactions, while maintaining alignment with the underlying data schemas. We further evaluate the system across multiple models and scenarios, focusing on the correctness of query generation, execution reliability, and the quality of derived insights.

Our results highlight both the effectiveness and the limitations of agentic approaches in real-world FinOps settings, emphasizing the importance of schema grounding and structured data access in improving reliability.