Abstract
Systems and methods are described for generating a tool catalog for conversational agents from historical conversation data, which can address challenges associated with manual catalog creation. The described technology can use a multi-stage process that may begin with a discovery engine, which could leverage a large language model, to analyze conversations and extract potential tools with their respective inputs and outputs. A subsequent refinement stage can employ a critic-actor architecture to iteratively identify and address potential logical issues in the extracted catalog, such as duplicate functions or unresolved parameter dependencies. This process can result in a dependency-aware tool catalog with improved logical consistency that may be used to aid the development and enhance the capabilities of virtual agents.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Huang, Joshua; Yuan, Steve; Horling, Bryan; and Zafarmand, Mahdi, "Automatic Discovery and Refinement of Tool Catalogs from Conversational Data", Technical Disclosure Commons, (September 24, 2026)
https://www.tdcommons.org/dpubs_series/11857