Gains:
- Ability to use AI efficiently in processing, quality control and reporting of laboratory results
- Ability to accelerate geological/geotechnical report drafting, literature review and data summary with AI
- Ability to secure automatically generated content with traceability, source confirmation and engineer approval
An invisible but large part of a geological engineer's job takes place in the office: processing laboratory results, quality control, tabulating data, writing reports, scanning literature, and preparing maps and figures. These jobs are repetitive, time-consuming and attention-demanding; This is exactly the area where artificial intelligence saves the most time. But there is a trap here too: the report is the official basis for an engineering decision. A fabricated source, a miscopied number, or an untraceable claim can slip into a flowing text unnoticed, and the consequences can be serious. The aim of this unit is to establish a discipline that never relaxes traceability, source confirmation and engineer approval while using artificial intelligence in end-to-end office efficiency.
In this unit we will cover three office fronts: laboratory result processing and QA/QC (cleaning and ensuring analysis data), report and literature production (drafting, abstracting and screening acceleration), and workflow automation (regularizing repetitive tasks). At every step, AI is a drafter and organizer; The accuracy, source and final approval of the content belong to the engineer.
Laboratory Result Processing and QA/QC
Laboratory results (grade analyses, soil index tests, water chemistry) often arrive in different formats, in different units and sometimes incorrectly. AI fits this data into a single schema, captures unit inconsistency, flags below the limit of detection and checks QA/QC samples (blank sample, standard reference material, reanalysis). QA/QC (quality assurance/quality control) is the inspection that proves the reliability of the laboratory: if there is contamination in the blank sample, deviation in the standard, non-compliance in the repeat sample, the data is questionable.
Critical point: AI never silently deletes or corrects a value. Its job is to flag problematic lines and let the engineer make a decision. An outlier may be a true high ore or a laboratory error; The engineer makes this distinction with the context. Artificial intelligence checks the data, does not clean it.
Tip: Repeat the "just mark, don't change" rule at every prompt when processing lab data. The most insidious mistake of AI is when it corrupts the raw data by "correcting" a value with good intentions and not realizing it.
Report and Literature Production
AI is a powerful drafter in report writing: drafts the method section, introductory context, conclusion summary in minutes; summarizes a long body of literature; Converts a complex table to plain text. But two dangers are evident. The first is the fabricated source: the model can confidently generate a non-existent article, a false author, or a non-existent canon article. Secondly, number handling error: a grade, a safety coefficient or a depth may be incorrectly transferred when passing into the text.
So the golden rule: AI produces draft, engineer verifies and owns it. Each source is confirmed from the original (apocryphal references are the most common type of hallucination); each number is compared to the raw data; Every technical claim is tested with engineering knowledge. Once the report is signed, the responsibility for every sentence in it lies with the signer — “AI wrote it” is not an excuse.
Attention: Never accept the bibliography produced by artificial intelligence as it is. Made-up but realistic-looking references (seemingly accurate author, journal, year) are the most common and dangerous hallucinations. Verify each reference against independent database.
Workflow Automation and Traceability
Automating repetitive tasks (processing monthly water monitoring data, filling out a standard report template, combining drilling tables) with artificial intelligence provides great efficiency. But automation can also automate error: a wrong rule once established is silently repeated at every run. That's why automated workflows must be traceable: recording which input, which step, which model version each output comes from; When an error is found, the root cause and all affected outputs must be traceable.
Traceability means in practice: the source and date of the input data, the transformations applied, the prompt/model version used and the human approval step are recorded. This ensures both debugging and engineering liability. Uncontrollable automation, no matter how fast, is unacceptable in a safety-critical business.
Three Mini Cases: By the Numbers
Case 1 — Capturing QA/QC drift. One laboratory shipment contained 320 grade samples. AI QA/QC screening flagged standard reference material as systematically 8% under-reading over the last 40 samples — instrument calibration drift. This systematic error was a problem that was very difficult to detect manually but affected the entire batch; The shipment was reanalyzed. He didn't decide the pattern, he made the pattern visible.
Case 2 — Fabricated source. AI drafted the literature section of a geotechnical report and produced six references. During verification, it was found that two of them never existed and one was attributed to the wrong journal. If the references had not been checked one by one against an independent database, the fabricated sources would have entered a signed engineering report. The draft sped up; Confirmation saved.
Case 3 — Root cause with traceability. In the monthly water monitoring automation, the values of an observation well were reported in the wrong unit for three months. Thanks to the traceability record, it was found within minutes at which conversion step the error started and a full list of the three affected reports was created. If no records were kept, it would remain unclear which printouts were corrupted and the entire archive would be suspect.
Weak Prompt / Strong Prompt
Weak prompt:
Write a geotechnical report for this project, include literature. [data]
Powerful prompt:
Your role: Report drafter. DRAFT a report with the following data but:1) Based ONLY on the data I give; DO NOT make up the number, source or finding. 2) Show with parentheses which input line each number comes from. 3) Instead of making up the actual source for the literature, put a placeholder "verified source about [topic] will be added here". 4) Mark anywhere you are unsure as "REQUIRE ENGINEER CONFIRMATION". The final content and all sources will be verified by me. The data is anonymous.
Four Copiable Templates
1) Laboratory QA/QC inspection:
Check laboratory results table: unit discrepancy, signs below detection limit, contamination in blank sample, systematic drift in standard reference, non-compliance in repeat sample. Only mark problematic rows with example_id; Do not delete or correct any values. The decision is mine.
2) Enforce source confirmation:
Pull out all numerical claims and source citations in this text and make a checklist: the input on which each number is based, the citation of each source that needs to be confirmed. Source fabrication; Mark "must be verified" where you cannot verify.
3) Traceability record framework:
Propose a traceability record template for this automated workflow: input source and date, transformations applied, prompt/model version used, human approval step. Show how to get to the root cause and affected outputs when an error is found.
4) Report consistency check:
Scan your report draft for internal consistency: do the numbers in the summary match the body, do tables and text conflict, are the units consistent throughout, are there undefined abbreviations? List the problems; The content decision is mine.
Office Duty, Earnings and Control
Quest
AI gain
Main risk
Mandatory check
Laboratory data processing
high
Silent correction
"Mark, delete" + raw data
QA/QC inspection
high
Don't miss the slide
Blind/standard/reconfirmation
Report draft
high
Made-up issue/source
Number and source confirmation
Literature review
high
Made-up reference
Independent database verification
workflow automation
very high
Repeat of mistake
Traceability record
Tip: Before setting up automation, ask "how will I know if this workflow is working incorrectly?" Answer the question. If the answer is not clear, add traceability and control points first; Speed comes after controllability.
Common mistakes
- Getting the value corrected silently. AI “improving” raw data is the most insidious mistake; the rule should always be "mark, don't change."
- Not verifying the bibliography. Made-up references are the most common hallucination; Each imprint must be independently verified.
- Not checking number carry. A tenor or safety coefficient that deteriorates when passing into the text is hidden in the fluent text.
- Establishing untraceable automation. In a workflow that does not keep records, the root cause and impact of the error cannot be determined.
- Taking refuge in the excuse that "AI wrote it". The responsibility for every sentence signed lies with the signer; approval cannot be relaxed.
In summary
- Office work is the area where artificial intelligence saves the most time; But the report is an official basis for decision and mistakes are expensive.
- In laboratory data, AI audits, not cleans; the rule is "mark, delete/correct", the decision is up to the engineer.
- Faking sources and numbers in reports and literature are the most common risks; Every source and number is verified.
- Automation can increase error as well as speed; The traceability record secures the root cause and affected outputs.
- The engineer is responsible for the signed content; "AI wrote it" is not an excuse.
Application task
Receive an anonymized lab results table (or representative data). Have the model flag problem lines with the "Laboratory QA/QC audit" template; consciously systematically shift a standard reference value and test whether the model catches it and tries to delete the value. Then, have a short report section drafted and every issue and source in it put into a checklist using the "source verification challenge" template.
checklist
- [ ] I always apply the “mark, delete/correct” rule to AI on lab data.
- [ ] I check laboratory reliability with QA/QC samples (blank, standard, repeat).
- [ ] I independently verify every source and number in the report and literature.
- [ ] I set up automation workflows with traceability.
- [ ] I accept that I have full responsibility for the content I sign.