Poliac: Production Observability for LLM-Based Infrastructure as Code

Abstract

Infrastructure as Code (IaC) enables the automated provisioning of cloud infrastructures, yet syntactically valid configurations may remain unsuitable for the applications they support. Addressing such issues requires knowledge of both the infrastructure and its runtime behavior. We introduce POLIAC, an LLM-based framework that combines an IaC artifact, a natural-language specification, and aggregated operational data to generate infrastructure recommendations grounded in the state of the deployed system. We conduct a preliminary study on CorrectExam, a web application deployed on Grid'5000, collecting system- and application-level metrics under load. We query four LLMs with and without these operational aggregates. Without operational data, the models propose different resource modifications; when operational data are provided, all models recommend increasing vCPU while preserving memory, consistently with the observed CPU saturation and moderate memory utilization. These preliminary results suggest that operational data can reduce ambiguity in LLM-generated IaC recommendations and help avoid unnecessary resource provisioning.