Inspected before it’s trusted.

On a government infrastructure project, nothing is accepted because it looks right. It is checked against criteria agreed in advance, released at hold points by a named person, and recorded. taliQ applies the same discipline to AI.

A string line pinned between points on a drafting board, connecting a sequence of paper notes.

Inspection discipline, applied to AI

taliQ’s founder is a senior civil and structural engineer in the NSW rail sector, with more than five years of client-side QA inspection on government infrastructure. These are six working tools from that job, and what each one becomes in an AI engagement.

  1. Inspection and test plan (ITP)

    On an infrastructure project
    Before work starts, the contractor and the client’s inspector agree a list of every check: what is inspected, the acceptance criterion, the reference standard, how often, and who signs.
    Applied to AI
    Before an AI use is trialled, the checks are written down: what output is checked, what counts as acceptable, how many examples, and who decides.
    What you receive
    Pilot inspection and test plan
    Where it fits
    • Guidance for AI Adoption, practice 5: Test and monitor
    • VAISS guardrail 4: Test AI models and systems to evaluate model performance and monitor the system once deployed.
  2. Hold point

    On an infrastructure project
    A stage the work cannot pass until a nominated person has inspected it and released it in writing. Pouring concrete before the reinforcement is released is a non-conformance, whatever the concrete looks like.
    Applied to AI
    No AI use goes live, and no customer information goes into a tool, until the business’s accountable owner has signed it off against the test evidence.
    What you receive
    Hold-point release record
    Where it fits
    • Guidance for AI Adoption, practice 1: Decide who is accountable
    • Guidance for AI Adoption, practice 6: Maintain human control
    • VAISS guardrail 1: Establish, implement and publish an accountability process including governance, internal capability and a strategy for regulatory compliance.
    • VAISS guardrail 5: Enable human control or intervention in an AI system to achieve meaningful human oversight across the life cycle.
  3. Witness point

    On an infrastructure project
    The client is told in advance and may attend the test. If they choose not to, the work can proceed and the result is still recorded.
    Applied to AI
    You are invited to watch the evaluation run on your own examples, so the result is not something you have to take on trust.
    What you receive
    Evaluation record
    Where it fits
    • Guidance for AI Adoption, practice 4: Share essential information
    • Guidance for AI Adoption, practice 5: Test and monitor
    • VAISS guardrail 4: Test AI models and systems to evaluate model performance and monitor the system once deployed.
    • VAISS guardrail 9: Keep and maintain records to allow third parties to assess compliance with guardrails.
  4. Evidence and records

    On an infrastructure project
    If it isn’t recorded, it wasn’t inspected. Test results, materials certificates and sign-offs are kept so a third party can check the work later.
    Applied to AI
    Every approved use, its tool and version, the test results and the release decision are kept in the AI-use register, so the decision can be checked by someone who wasn’t there.
    What you receive
    AI-use register
    Where it fits
    • Guidance for AI Adoption, practice 4: Share essential information
    • VAISS guardrail 3: Protect AI systems, and implement data governance measures to manage data quality and provenance.
    • VAISS guardrail 9: Keep and maintain records to allow third parties to assess compliance with guardrails.
  5. Non-conformance report (NCR)

    On an infrastructure project
    When work does not meet the specification it is recorded, a disposition is decided (accept as is, rework or reject), the root cause is found and a corrective action is closed out.
    Applied to AI
    When an AI output is wrong in use (a wrong figure sent to a customer, or private information pasted into a public tool), it is logged, dealt with, traced to its cause and the prompt, data, rule or training is fixed.
    What you receive
    Non-conformance report
    Where it fits
    • Guidance for AI Adoption, practice 3: Measure and manage risks
    • Guidance for AI Adoption, practice 5: Test and monitor
    • VAISS guardrail 2: Establish and implement a risk management process to identify and mitigate risks.
    • VAISS guardrail 7: Establish processes for people impacted by AI systems to challenge use or outcomes.
  6. Client-side independence

    On an infrastructure project
    The inspector works for the asset owner, not the contractor, and is paid to find problems rather than to get the work accepted.
    Applied to AI
    taliQ is engaged by you, not by a tool vendor, and will put a stop or fix-first decision in writing when that is the honest answer.
    What you receive
    Written go, fix-first or stop decisions
    Where it fits
    • Guidance for AI Adoption, practice 2: Understand impacts and plan accordingly
    • VAISS guardrail 8: Be transparent with other organisations across the AI supply chain about data, models and systems to help them effectively address risks.
    • VAISS guardrail 10: Engage your stakeholders and evaluate their needs and circumstances, with a focus on safety, diversity, inclusion and fairness.

What happens, and what you get, each week

The path below combines the two AI offers. Each offer uses only the weeks marked for it. Times are the client’s time, not taliQ’s.

The frameworks behind the artefacts

All are voluntary. Existing Australian law, including the Privacy Act 1988 (Cth) and Australian Consumer Law, still applies to AI use. taliQ does not give legal advice.

See the artefacts themselves: Sample AI acceptable-use policy, Use-case assessment worksheet and the worked scenario.

What would you like to work on?

Bring the question you have now. The first call is free.