AI policy

1What we use AI for

• Reading drawings: sorting the sheets of a drawing set by what they show, and naming the rooms and equipment on a plan.
• Reading standards: drafting checkable rules from your guideline PDFs, each quoting the clause it came from.
• Judging rules: deciding whether a drawing meets each rule, given the measurements our code has taken.
• Writing: report narrative, suggested fixes, notes on recommended layouts, and answers from the in-app assistant.

2What we never use AI for

• Measuring. Clearances, counts, placements and room sizes are computed from the drawing’s geometry in our own code. The model is handed those numbers, never asked for them.
• Scoring. The compliance score follows from the findings by a fixed formula, so the same drawing always scores the same.
• Ranking layouts. Approved layouts are ranked by size, shape and equipment mix in code. The AI only writes the notes on each option.
• Sizing capacity. Freezer, chiller and dry store requirements come from your organisation’s calculator, not the model.
• Deciding. The AI never approves, rejects or sends back a store.

3A person signs off every decision

• Rules drafted by the AI are reviewed, edited and published by your staff, who also choose which rule set version each check runs against.
• A reviewer corrects the equipment and rooms the AI found and confirms them before any check runs. Items left unclassified are excluded from checks.
• A submitted revision needs approval from both your design reviewer and your QA/Ops reviewer. Either can send it back.

4Sources and doubt are shown

Every rule quotes the sentence of the guideline it came from, and every finding names the rule and clause it relies on. Where the AI is unsure equipment is missing, the finding reads “possibly missing” rather than a failure, and where the drawing cannot prove something, the report says so instead of guessing.

5Your data and model training

AI steps run on a third-party model provider, called through its API. Under the provider’s API terms, data sent this way is not used to train its models by default. We do not train models on your drawings, standards or reports, and we do not sell them. The provider is named, with our other subprocessors, in the privacy policy at https://niyam.nirmaanx.in/privacy.

6Enterprise

On the Enterprise plan, the AI steps can run on your own model account, such as Azure OpenAI or Amazon Bedrock, under your own agreement with that provider. We also provide an AI governance pack for your risk team describing each AI step, its inputs and its outputs.

7Records

Every check run, rule set version, submission, approval and send-back is stored with its time, so any finding can be traced to the rules and drawing it was made from.

8Changes to this policy

We may update this policy as the product changes. We will change the date at the top, and tell your admins in advance before we use AI for anything this policy says we don’t.

9Contact

Questions about how NiyamX uses AI can be sent to info@nirmaanx.in or +91-70411-80305.