Applying Action Research to Developing a GPT-Based Assistant for Construction Cost Code Verification in State-Funded Projects in Vietnam

Cost code verification in state-funded construction projects remains a labour-intensive and error-prone task, particularly given the structural heterogeneity of project estimates and the prevalence of malformed codes, inconsistent UoMs, and locally modified price components. This study evaluates a deterministic GPT-based assistant designed to automate Vietnam’s regulatory verification. The system enforces strict rule sequencing and dataset grounding via Python-governed computations. Rather than relying on probabilistic or semantic reasoning, the system performs strictly deterministic checks on code validity, UoM alignment, and MTR - LBR - MCR conformity with provincial Unit Price Books (UPBs). A dedicated exact-match mechanism, activated only when a code is invalid, enables the recovery of typographical errors solely when a project item’s full price vector matches a normative entry exactly. Using twenty real construction estimates (16,100 rows) and twelve controlled error-injection cases, the study demonstrates that the assistant executes verification steps with high reliability across diverse spreadsheet structures, avoiding hallucination and maintaining full auditability. Deterministic extraction and normalisation routines facilitate robust handling of displaced headers, merged cells, and non-standard labelling, while structured reporting provides line-by-line traceability aligned with professional verification workflows. Practitioner feedback confirms that the system reduces manual tracing effort, improves inter-evaluator consistency, and supports compliance documentation without encroaching on human judgment. This research contributes a framework for LLM-orchestrated verification, demonstrating how Action Research can align AI tools with domain expectations. Furthermore, it establishes a methodology for deploying LLMs in safety-critical, regulation-driven environments. Limitations—including narrow diagnostic scope, TT quotation exclusion, single-province UPB dependence, and sensitivity to extreme spreadsheet irregularities—define directions for future deterministic extensions. Overall, the findings illustrate how tightly constrained LLM configurations can augment, rather than replace, professional cost-verification practice in public-sector construction.