A defensible classification, in minutes — with the rationale on every output.

Why classification goes wrong

Most validation overspend starts here. A configured commercial product treated as a custom application can double the document set and add months, and the mistake is usually caution rather than analysis — nobody felt confident enough to argue for the lower category, so the project paid for the higher one.

The opposite error is quieter and more serious. A heavily configured system labelled Category 3 produces a thin package that looks efficient until an inspector asks why the configuration was never specified or tested.

The other failure is treating everything as software. An HPLC is not a GAMP category; a bioreactor is not a GAMP category. Forcing instruments and equipment through a software lifecycle produces documents that do not match what a qualification auditor expects to see.

One intake, four tracks

You give the source document — a URS, a BRD or a vendor specification. The platform establishes the facts about the asset, shows you what it concluded and the text it relied on, and you confirm or correct. Correcting a fact re-derives everything downstream rather than leaving a stale classification behind.

Application software and infrastructure follow computerised system validation and receive a GAMP 5 category. Analytical instruments follow qualification and receive a USP <1058> group. Equipment follows qualification against EU GMP Annex 15 criticality. Where an instrument has a software component — an HPLC with a chromatography data system — both are handled in the same project, each under the right standard.

Computed, not generated

The classification itself is calculated from the confirmed facts. The AI answers factual questions about the asset and quotes where in your document it found each answer; it does not decide the category. That distinction matters when an inspector asks how a system came to be classified, because the answer is a rule applied to stated facts rather than a model's opinion.

It also means the result is reproducible. Run the same facts through twice and you get the same classification, which is not something you can say of a generative answer.

The decision trail

Every classification carries its basis: which facts drove it, what text they came from, who confirmed them and when. If a fact is later corrected, the change is recorded and the downstream documents are re-derived rather than silently diverging.

This is what turns classification from an opinion written in a plan into something you can defend two years later when nobody involved still works there.