Research Article

Dynamic Production Issue Triage Through Declarative Workflows

Abstract

Production support teams spend a disproportionate amount of time on repetitive, crossservice triage: correlating order state, provisioning records, logs, and ticket history to answer why a given transaction is broken. Existing runbook automation is typically hardcoded, owned by a single engineering team, and brittle to new failure modes. We present the Workflow Intent Graph System, an architecture that lets any team declare a triage or operational workflow in YAML, compiles that declarative definition into a LangGraph state graph, and uses a large language model (LLM) to match natural-language queries directly to the graph node representing the correct entry point, then navigates the graph at run time to reach a diagnosis or an automated fix. The system supports sub-workflow reuse as edges between independently authored subgraphs, parallel step execution, and dynamic multi-workflow chaining, without requiring workflow authors to write orchestration code. We describe the architecture, the YAML-to-LangGraph compilation step, the node-level intent-resolution pipeline, and the resulting extensibility, auditability, and performance characteristics.

Authors affiliation:

Kumar Chandan: No. 422 Deerhill Dr, San Ramon, CA -94583, USA

How To Cite: K. Chandan. Dynamic Production Issue Triage Through Declarative Workflows. Journal of Computers and Applications, 2(2):48–53, 2026. https://doi.org/10.58613/jca225