Integrating AI into SRE Workflows — Part 1: Automating Infrastructure Cost Estimation with Smart RI Calc

The Yeogi Eottae Company SRE team developed Smart RI Calc to automate infrastructure cost estimation and EKS node scaling predictions, which were previously repetitive manual tasks. The existing approach involved manually summing RI Excel price sheets and performing manual calculations via SSH access, taking 20 minutes per project and resulting in discrepancies between engineers.

As a solution, the team established the principle: 'AI reads, code calculates, humans review.' The LLM extracts resource information from unstructured Confluence meeting notes, while backend code uses the AWS Pricing API to calculate exact unit prices. The final results are verified by humans through a Human-in-the-loop structure to prevent errors.

In particular, EKS node scaling predictions based on Karpenter were implemented as simulations reflecting actual operational data, rather than simple scheduler replication. The initial model deviated from actual deployments due to over-quoting and the omission of topologySpreadConstraints, but accuracy was improved by applying a correction formula based on minimum availability criteria and a rule prohibiting majority occupancy. This reduced estimation time and enabled consistent results.