Enterprise software often exposes functionality through relational databases and REST APIs, but using them typically requires knowing the underlying schemas, endpoints, and parameters. This creates a barrier for users who want to interact with systems in plain language. The new arXiv paper, a revised preprint, addresses this by exploring retrieval-augmented generation (RAG) as a way to bridge the gap.

The paper outlines RAG strategies for two related tasks: generating SQL queries from natural language and producing REST API calls. Rather than relying on the model to memorize every schema or endpoint, RAG retrieves relevant context at generation time. This approach is meant to make the generated queries and calls more accurate and adaptable to specific enterprise environments.

Because the source is a single preprint, the findings should be treated as early-stage rather than conclusive. Still, the work points to a practical direction: using retrieval to supply the technical details that users should not have to memorize, potentially lowering the barrier to interacting with complex enterprise systems.