Generating Synthetic Ansible Programs with Probabilistic Methods

In this talk we will take a high level look at a possible approach for generating synthetic values for Ansible parameters. These synthetic values would be used in templated Ansible programs and be integrated into real world datasets of Ansible Programs. These would serve as valuable resources for training Large Language Models (LLMs) on the Ansible programing language. As of now, our research on this problem is promising but not conclusive and TUW will further collaborate with IBM on this subject.