This project demonstrates a structured, ChatGPT-assisted approach to reviewing engineering maintenance documentation. The review uses a fictional six-step centrifugal pump maintenance procedure for equipment P-101.
The objective is to identify documentation gaps, ambiguous instructions, missing acceptance criteria, safety-related omissions, and items requiring further verification.
Project type: AI-assisted engineering document QA case study
Domain: Mechanical Engineering, Industrial Maintenance, Technical Documentation
Tools: ChatGPT, GitHub, structured review prompts, PDF reports
- Identify missing information and ambiguous maintenance instructions.
- Trace findings to the relevant source-procedure steps.
- Recommend corrective actions and appropriate verification sources.
- Distinguish documentation gaps from confirmed real-world hazards.
- Compare a baseline review with a review using more explicit instructions.
- Document limitations and the need for qualified human verification.
A fresh review was prepared using the fictional maintenance procedure and a structured prompt. The report identifies documentation gaps and provides a findings register, checklist, unanswered questions, and safety limitations.
A second review used more explicit instructions for finding traceability, priority rationale, confidence ratings, non-duplication, and verification planning.
The two available reports were compared to examine differences in finding coverage, structure, uncertainty handling, and verification planning. The comparison records observed differences rather than assuming that a higher finding count automatically means better quality.
Execution transparency: The reports were prepared in separate ChatGPT conversations with human-directed prompts. This repository documents a ChatGPT-assisted iterative review workflow. It does not demonstrate an independently deployed or autonomous AI agent, and no separate agent-platform trajectory has been captured.
- Run 1: 9 preliminary documentation findings.
- Run 2: 13 preliminary documentation findings.
- Run 2 adds more explicit priority rationales and confidence ratings.
- Run 2 provides a more structured verification plan and checklist.
- The comparison highlights the importance of traceability, uncertainty handling, and human review.
Finding counts are descriptive, not a standalone measure of quality. The findings require technical review to assess accuracy, overlap, and practical relevance.
| File | Description |
|---|---|
Run_1_Engineering_Document_QA_Report.pdf |
Fresh baseline review report |
Run_2_Engineering_Maintenance_Procedure_Quality_Review.pdf |
Review using improved instructions |
Run_1_vs_Run_2_Engineering_Document_QA_Comparison.pdf |
Updated comparison of the two review outputs |
Run_1_vs_Run_2_Comparison_and_Final_Capstone_Report.pdf |
Earlier comparison and capstone report |
The reviews consider:
- Energy isolation and verification requirements
- Safe access and risk-assessment references
- PPE requirements and applicable procedures
- Inspection methods and acceptance criteria
- Component damage and replacement criteria
- Reassembly, pre-start checks, and restart authorization
- Post-start monitoring and maintenance records
- Document control, unresolved questions, and verification needs
- The maintenance procedure is fictional and intended for training.
- The findings have not been independently validated against an OEM manual, site procedure, engineering drawing, or formal risk assessment.
- Priorities are preliminary reviewer judgments, not formal risk ratings.
- A gap in the supplied excerpt does not prove that a control is absent from the wider site maintenance system.
- The reports must not be used as operational maintenance instructions or as evidence of procedure approval.
- Qualified personnel must verify engineering and safety requirements before real-world use.
- Engineering documentation review
- Technical writing and quality assurance
- Prompt design and instruction refinement
- Evidence traceability and structured reporting
- Comparative output evaluation
- Identification of uncertainty and verification requirements
- Safety-aware human oversight
Summary: This case study illustrates how structured prompts and human review can support engineering document quality assurance. It emphasizes traceable findings, careful handling of uncertainty, and transparent reporting of limitations.