Agents author.
Your engineers approve.

One chain of custody, from requirement to signed report

Validation fails audits at the joins - where a requirement is re-typed into a test script, or evidence is pasted without its source. Qualitum removes the joins. Each artefact is generated from the one before it and keeps a live link back to the page it came from.

The project workspace holds every document the equipment will ever be judged against

Drop in the URS, the vendor quotations, mechanical and functional specs, layouts, alarm lists, P&IDs and signed change orders. Qualitum reads each one, identifies what it is, and makes it addressable at chapter and page level.

Why it matters: validation teams lose weeks re-finding the document that justified a decision. Here the library is the evidence base - nothing is cited that is not in it, and nothing in it is invisible to the engine.

The URS becomes 247 individually addressable requirements - with their source still attached

Qualitum splits the specification into atomic, testable requirements, assigns a stable ID, and records exactly where each one came from: section, page, and the sentence itself. GxP relevance and criticality are proposed at the same time.

Why it matters: this is the step teams normally do by hand in Word and Excel over several weeks. Doing it once, correctly, with provenance intact, is what makes every later stage automatable.

Open any requirement and you land on the sentence that created it

The detail panel names the root document, the revision, the chapter and the page, and highlights the exact wording in the original PDF beside it. Nothing is paraphrased silently: if the platform rewrote a sentence to make it testable, both versions are shown.

Why it matters: in an inspection the question is never "what does the system say" - it is "show me where that came from". This panel is the answer, in one click, without leaving the tool.

Criticality decides how hard a requirement has to be tested

Each requirement is classified for GxP impact, patient risk, data integrity and detectability. Qualitum proposes the classification with its reasoning, the quality lead confirms or overrides it, and the resulting test strategy - which protocols the requirement will appear in - follows automatically.

Why it matters: risk-based validation is expected by GAMP 5 and by every inspector, but it is usually argued in a spreadsheet nobody can reconstruct later. Here the argument, the decision and the consequence sit on the same record.

Every requirement, checked against every design document, in one grid

The traceability matrix runs each requirement against each document in the project and returns the chapters that actually answer it - with chapter number, page and a supporting quote. Where the engine finds several plausible answers, it presents them as candidates rather than guessing.

Why it matters: this is the work that normally takes a senior engineer four to six weeks per system, document by document. Qualitum does the reading; the engineer does the deciding.

Where the engine is not certain, it asks - and shows its working

Each candidate carries its chapter name, chapter number, supporting quote, page and confidence. The reviewer selects, edits or adds a candidate; the decision, the reviewer and the timestamp go to the audit trail. Confidence is never used to auto-approve a critical requirement.

Why it matters: an automated matrix nobody trusts is worse than none at all. Presenting candidates with evidence turns review into a minute-long confirmation instead of an hour-long search - and keeps a qualified human accountable for the decision.

Vendor offers compared requirement by requirement, not chapter by chapter

Load two or more supplier quotations and Qualitum answers the only question that matters: for each of your 247 requirements, does this offer meet it - fully, partially, or not at all? Every verdict carries the supplier's own wording and the page it appears on.

Why it matters: technical bid evaluation is normally a subjective read of two long PDFs. This turns it into a defensible, itemised comparison you can put in front of procurement, quality and the vendor at the same time.

A confidence score you can interrogate, next to the page it came from

Confidence is not decoration. It reflects how directly the source text answers the requirement, whether the wording is verbatim or inferred, and whether other passages in the same document contradict it. Open the score and the platform highlights the passage in the supplier's own PDF.

Why it matters: reviewers need to know where to spend attention. High-confidence verbatim matches are confirmed in seconds; anything below threshold is routed to a human before it can reach a protocol or a report.

Coverage, confidence and open risk on one page - at any hour of the project

Validation status is normally a monthly slide built by hand. Here it is a live view: how much of the URS is evidenced, how much is tested, where confidence is thin, what is blocking, and who owns the next action.

Why it matters: the head of validation can answer "are we ready for FAT?" with a number and a list rather than an opinion - and can show an inspector that the open items were known and managed.

IQ, OQ and PQ protocols drafted from the matrix - not from a template

Every approved requirement produces test steps written against its own acceptance criteria, in your company's protocol format. Each step keeps the URS ID, the design evidence it was derived from, and the reason it exists. Steps with no requirement behind them cannot be created.

Why it matters: orphan tests and untested requirements are the two findings inspectors reach for first. When protocols are generated from the traceability matrix, neither can occur - and the drafting week disappears.

FAT and SAT executed in the system, with deviations raised where they happen

Testers record the observed result against the acceptance criteria, attach photographs or instrument printouts, and sign. A failing step raises a deviation on the spot, links it to the requirement, and blocks the protocol from closing until it is resolved or justified.

Why it matters: the gap between a marked-up paper protocol and a final report is where weeks and credibility are lost. Executing in place means the report is already written when the last step is signed.

Change one requirement and see everything it touches, before you commit

A revised URS clause, a design change, a new supplier revision - Qualitum runs the impact across the whole chain and lists the design evidence, test steps, executed results and signatures affected. Approve the change and the addendum protocol is drafted with only the delta re-tested.

Why it matters: re-validation cost is driven by not knowing the blast radius, so teams re-test everything. Knowing exactly which twelve steps are affected turns a six-week addendum into a three-day one.

Every action attributable, every signature meaningful

Qualitum records who did what, when, from where, and what the value was before and after - including every action taken by the AI engine, named as such. Signatures carry the signer's printed name, the meaning of the signature and the record it applies to, and cannot be transferred or reused.

Why it matters: 21 CFR Part 11 and EU GMP Annex 11 do not care that a system is intelligent; they care that it is attributable, contemporaneous and unalterable. Automation only counts if its own actions are logged like anyone else's.

The deliverables come out in your format, already reconciled

Requirement traceability matrix, DQ report, IQ/OQ/PQ protocols and summary reports, deviation log, vendor comparison and clarification list - generated from the live data in Word, Excel or PDF, in the customer's own templates, with citations preserved as footnotes.

Why it matters: the document is still the regulatory deliverable. Qualitum does not ask a quality organisation to change how it files - it removes the two weeks of copying, cross-checking and renumbering that produce the file.

Five controls stand between a model output and a validation record

Nothing reaches a protocol on a single model pass. Each extraction and each match is sampled several times, checked against the source text, scored, and routed to a human where it is thin or where criticality demands it - regardless of score.

Why it matters: this is what separates an assistant from a validated system. It is also what lets Qualitum run inside your own environment, on your choice of model, without your data leaving it.