The Sovereign profile of Tonia runs technical-specification drafting and plan analysis on-site installed in your office: project data (financial plans, strategic schedules, the client's industrial secrets) execute locally, never cross-border. The Frontier profile routes non-identifying plan analysis to a frontier-model provider with redaction of company names and project addresses. The locally signed audit log documents AI use in practice — the tool the OIQ asks for to demonstrate professional diligence.
The consulting-engineering firm — whether in structure, geotechnics, environment, building mechanics or electrical — operates under three simultaneous regimes: Bill 25, the Engineers' Code of professional conduct, and the Engineers Act (CQLR, c. I-9) with its professional-liability duties.
The practical consequence is known to every firm that has experimented with ChatGPT for specification drafting since 2023: pasting a client brief ("our client, owner of a CA$45M industrial complex in Saint-Bruno, plans an aerospace expansion") into ChatGPT communicates professional secrecy on the client's expansion strategy to a U.S. corporation, without consent, without a contractual framework, and — if AI error contaminates the signed deliverable — engages the engineer's civil liability.
(drafting the technical portion of a tender, structuring an equipment specification). Case where AI delivers most value in time reduction. But the technical specification often carries confidential project data (precise site address, client contract amount, strategic schedule, identified partners). Pasting the real brief into ChatGPT exposes both client and firm.
(spec extraction, dimensional verification, comparison with prevailing standards). Low sensitivity in itself if the plans do not identify the client. Still problematic if the plan title block carries the owner's name, the project address, or unique internal references.
(post-incident analysis, judicial expertise for construction litigation, collapse-cause assessment). Sensitive: the report may reveal litigious elements that will end up in court. The expert report is an engineer-signed deliverable; its content engages professional liability.
The on-site tonia installed in the office absorbs the drafting of specifications carrying confidential project data. The client brief runs through the local model; no project data leaves the office. The signed audit log documents the AI contribution for OIQ traceability.
The Frontier profile routes approved requests to a frontier-model provider. tonia redacts company names, project addresses, unique internal references (mandate numbers, client codes) before send. Analysis happens on the redacted content; results return to the office; the engineer applies the missing client context on review.
Strictest posture. An expert report destined for litigation must be defensible in court; any AI contribution must be traceable and reproducible. The on-site tonia keeps the draft, the assisted writing, and the final expertise on-premises. The locally signed audit log becomes the evidence element demonstrating the signing engineer's diligence.
Mid-market consulting-engineering firm in Estrie, 40 employees (28 engineers + 12 support), specialized in structure and geotechnics, ~120 active mandates, deployed under Tonia — Sovereign profile in Q2-2026. Anonymization required for publication. OIQ validation referenced.
The firm had been experimenting with ChatGPT-4 for the technical portion of tender drafting since 2023. The 2025 publication of the OIQ Position on generative AI and an adjacent dispute in the milieu (a colleague sanctioned by the syndic for unsupervised AI use in an expert report) triggered an internal review. April audit: 4 categories non-compliant with art. 17, and a potential deontological flag on the traceability of AI contributions in already-delivered expert reports.
Switch in Q2-2026: on-site tonia in the office's server room, capacity sized for simultaneous drafting of 6 technical specifications at peak. Policy configured by the PRPRP with the managing partner. 2-h training of the 28 engineers + 1-h training of support staff. Creation of an internal procedure: every AI-assisted deliverable must name the AI contribution and attach the corresponding audit-log entry to the mandate file.
BLOCK event count (sectoral PII patterns — corporate client names, project addresses)Want to see how this applies to your office? Start with the free Loi 25 audit, then request a 30-min consultation. We will review your three use cases, OIQ obligations, and on-site tonia sizing if your context calls for it.
Disclosure notice: this page is editorial and reflects Tonia's commercial position. Regulatory references are verifiable at the indicated links. Before acting, validate the obligations specific to your organization with your counsel.