Left to right. URS to retirement. One trace graph.
The GAMP 5 V-model is the spine of every validation engagement. Validate·AI runs every stage as an agent-authored, human-approved workflow - and keeps the requirements traceability matrix live throughout. No orphan requirements. No stale links.
User Requirements
Decomposed from stakeholder inputs into testable requirements with risk tags.
Functional Spec
Drafted from URS, structured to your template; reviewer-assistive.
Design Qualified
Design qualified against URS with continuous trace links to FS and tests.
Build / Configure
Configuration captured, changes traced, deviations flagged at source.
Installation Q
Drafted from P&ID, OEM manuals, and your SOPs. Test scripts pre-populated.
Operational Q
Functional test execution against acceptance criteria. Deviations triaged inline.
Performance Q
Performance verified across operating ranges. Continuous monitoring post go-live.
Operational use
System in production. Live audit trail. ALCOA+ checked on every cycle.
Periodic review
Triggered by data, not by calendar. Revalidation scope auto-scoped.
Retirement
Data migrated under control; records retained per 21 CFR Part 11.
Five capabilities. One trace graph.
The cross-cutting work the agent runs across every validation domain. URS in. Audit trail review out. The trace graph in between.
Match candidate equipment to your URS - defensibly.
When validation has to bless an equipment selection, the agent scores each candidate vendor's specs against your URS, surfaces gaps, and drafts the DQ. Procurement runs in your ERP - SAP, Oracle, or Workday. The technical defensibility runs here.
The agent ingests OEM data sheets, FDS, vendor proposals, and your URS. It decomposes the URS into testable criteria, extracts vendor specs per criterion, runs pairwise scoring with confidence, and surfaces gaps. The DQ draft writes back to Validate·AI ready for QA review. Auditor-defensible, vendor-neutral.
No PO data leaves your perimeter. No commercial terms flow through Qualitum. The technical decision is documented; the commercial one stays in your ERP.
One validated system, many jurisdictions.
Global pharma operates the same system across countries, sites, and equipment generations - each with its own validated state. Validate·AI's trace graph tracks every variant: which site (Cork, Basel, Mumbai), which country regulator (FDA, EMA, MHRA, PMDA, ANVISA, CDSCO), which equipment serial number, which SOP revision.
Periodic review triggers fire per jurisdiction. A change at the Basel site doesn't drag the Mumbai site into revalidation unless the trace graph says it must. Inspection readiness is per-jurisdiction: when the PMDA inspector arrives at the Tokyo facility, the platform serves the Tokyo-specific defence pack.
No spreadsheets. No parallel matrices. No drift between “what's validated where.” The trace graph is the single source of truth, and it is queryable.
The traceability matrix is the system. Not a quarterly deliverable.
In every legacy validation platform, the RTM is a controlled document. Built at the start of the project, frozen, signed, archived - then rebuilt from scratch when anything changes. Engineers spend more time maintaining the RTM than authoring tests. By the time it's signed, it's stale.
The RTM is not generated. It is the underlying data model. Every requirement, every test step, every signature, every deviation, every change record - all stored as connected objects in a trace graph. Add a requirement, the graph extends. Update a test script, the graph re-resolves. Sign an executed protocol, the trace solidifies. PDF export only when the auditor wants paper.
The RTM is live, queryable, and navigable in real time. “Show me the test evidence for URS-014 across the Cork site.” Two clicks. “Show me every requirement that hasn't been tested in the last 12 months.” One query. Kneat, ValGenesis, and Veeva ship the document. Qualitum ships the data.
100% audit trail review, every cycle. Not 1% sampling.
MHRA Data Integrity Guidance (2018) §6.13 expects routine audit trail review proportionate to risk. WHO TRS 1019 Annex 5 says the same. PIC/S PI 041-1 §9 says the same. What inspectors actually find is sampling at 1–2 percent, justified by headcount. Anything beyond that is impossible at human throughput.
Every record passes the nine ALCOA+ criteria - attributable, legible, contemporaneous, original, accurate, complete, consistent, enduring, available - on every cycle. Validation happens twice: at write time (the moment of capture) and at review time (against the persisted record, independently). Drift between the two surfaces as a deviation draft for QA. Sample sizes regulators expect, at throughputs humans cannot sustain.
A continuous audit trail review record per system, per period, citable to the underlying data-integrity standard. When the MHRA inspector asks “show me your audit trail review for the last quarter on the Cork MES”, you serve the current revision - signed, with the deviations triaged, the CAPAs linked, and the trend chart of where you've improved.
Every validation domain. One platform.
Validate·AI runs across the full pharma validation lifecycle. Some domains are deep enough to deserve their own page. The rest are part of the same agent, applying the same capabilities across the same trace graph.
User Requirements Specification
Decomposed from stakeholder inputs into testable, risk-tagged requirement objects.
Functional & Design Specification
Drafted from URS, structured to your house template; reviewer-assistive.
Design Qualification
Candidate equipment scored against URS. DQ drafted - vendor-neutral, defensible.
Risk assessment
GAMP 5 categorisation, FMEA, ICH Q9(R1) applied per requirement and per system.
Installation Qualification
Drafted from P&ID and equipment manuals. Test scripts pre-populated.
Operational Qualification
Functional execution against acceptance criteria. Deviations triaged inline.
Performance Qualification
Performance verified across operating ranges. Continuous monitoring post go-live.
Factory & Site Acceptance Tests
Vendor-side and site-side test packs with executable scripts and signed evidence.
CSV / Computer Software Assurance
Risk-first testing. Critical thinking documented. CSA-aligned per FDA Sep-2022 draft.
Cleaning validation
PDE/ADE, MACO across the equipment train, swab recovery, three confirmatory runs.
Method validation & AIQ
ICH Q2(R2). USP <1058> Group A/B/C classified, qualified, method-validated.
Cold-chain & thermal mapping
WHO TRS 961. Probe placement designed, data aggregated, conclusions drafted.
Process validation
FDA 2011 lifecycle. Stage 1 / 2 / 3. PPQ, CPV. ICH Q8/Q9/Q10 aligned.
Sterilisation & aseptic
Annex 1 (2022). Media fill, PUPSIT, contamination control strategy.
Cleanroom & utilities
HVAC, WFI, PW, Pure Steam, CDA. ISO 14644-1. Requalified by data, not calendar.
Periodic review & revalidation
Triggers fire on the trace graph. Revalidation scope scoped, not blanket.
Requirements Traceability Matrix
Live trace graph between URS, FS, DS, test cases, deviations, and signed records.
Audit trail review (ALCOA+)
Every record, every cycle. 100 percent, not sampled. Findings drafted as deviations.
Validation Master Plan
Annex 15 VMP authored and maintained as data, not as a frozen Word file.
Inspection readiness
FDA, EMA, MHRA, PMDA, ANVISA. Defence pack and walkthrough generated on demand.
What gets shorter. What gets sharper.
Six measurable outcomes from production deployments. None of these are projections.
Authoring time saved
Reduction across IQ/OQ/PQ protocols, FAT/SAT, and CSV/CSA packages.
Review cycles
From weeks to days. Reviewer-assistive draft means QA review converges.
Right-first-time on test
30 to 50 percent lift on test execution; deviations caught at draft, not in protocol.
Deviation triage
Minutes, not days. Agent drafts root cause and CAPA; QA owns the signature.
RTM completeness
Continuous. No orphan requirements. Live impact analysis on change.
To first audited agent
From kickoff to production - 8 to 12 weeks for first audited agent.
The questions a Head of Validation actually asks.
Real questions, real answers. No marketing fluff. The standard is cited when it matters.
Yes. The agent classifies each system per GAMP 5 Second Edition - Category 3 (non-configured COTS), Category 4 (configured), Category 5 (custom developed). Classification drives test depth, supplier assessment scope, and lifecycle artefact requirements. Category 1 (infrastructure) and the supplier/internal split are also handled.
Classification is reviewer-assistive, never autonomous. The agent proposes the category and the reasoning. Your CSV lead confirms, escalates, or overrides. The categorisation decision and rationale are part of the audit trail.
CSA is native. Test scripts are scoped against patient-safety, product-quality, and data-integrity risk - not against a defaults checklist. Low-risk functionality gets unscripted dynamic testing or scripted record-light testing. High-risk functionality gets the full scripted approach with formal evidence capture.
The FDA Draft Guidance "Computer Software Assurance for Production and Quality System Software" (September 2022) is operationalised. ISPE GAMP 5 Second Edition (July 2022) and the ISPE GAMP CSA Concept Paper align here.
Electronic signatures (§11.50, §11.70, §11.100, §11.200). Each signature captures the printed name of the signer, date and time of signing, and the meaning of the signing. Signatures are linked to their records such that they cannot be excised, copied, or transferred. Two-component signing (identification + secret) with biometric option supported.
Audit trails (§11.10(e)). Computer-generated, time-stamped, independent of operator action. Records the operator, the date/time, the action, and the prior value. Tamper-evident at the platform layer.
Record retention (§11.10(c)). Records protected to enable their accurate and ready retrieval throughout the retention period. Configurable retention windows per record class; legal-hold supported.
§4 (Validation). The platform itself is validated as a GAMP 5 Category 4 baseline; site-specific configuration is validated as part of the customer's CSV lifecycle. Validation Summary Report available under MNDA.
§7 (Data storage). Data resides inside your tenant, encrypted at rest with customer-managed keys. Periodic checks for accessibility, readability, and accuracy are scheduled and evidenced.
§9 (Audit trails). Computer-generated, time-sequenced, tamper-evident. Recorded with sufficient detail to reconstruct the sequence of activities. Reviewed continuously - not sampled.
§17 (Archiving). Records archived in a manner that preserves data integrity for the regulatory retention period. Migration controls maintain readability across platform upgrades.
Live. The RTM is not a document - it is a graph. Every requirement (URS), every specification (FS, DS), every test case (IQ, OQ, PQ), every signed record is a node. Every relationship is an edge. The agent maintains the edges continuously as artefacts evolve.
Impact analysis on change is therefore a graph traversal, not a manual audit. When a URS clause changes, the platform tells you exactly which FS items, test cases, and signed records are affected - immediately.
The RTM can be exported as a document for inspection submission. The source of truth is the graph.
The agent owns triage. Your QA owns the decision and the signature.
When a test fails, the agent immediately drafts a deviation record - what failed, what was expected, what was observed, what the immediate impact assessment looks like, what the proposed root cause is (with citations to the URS clause, SOP, and prior CAPAs). The draft is logged in the deviation queue.
The investigator reviews the draft. Adjusts root cause if needed. Approves or rejects the proposed CAPA. The signature is human, attributable, and part of the audit trail. Handoff to your eQMS (TrackWise, Veeva Quality, MasterControl, ETQ) is bi-directional.
Every record passes ALCOA+ checks on every cycle. Not sampled. Reviewed.
Attributable - every action carries the authenticated identity of the actor (human or agent). Legible - records render to the standard your QA uses. Contemporaneous - timestamps are captured at the moment of the action, by a server clock, not the client. Original - the source record is preserved; copies are explicitly marked as such. Accurate - validation checks at write time, audit checks at review time.
The plus-four (Complete, Consistent, Enduring, Available) is enforced as part of the platform's continuous audit pipeline. MHRA's "GxP Data Integrity Guidance and Definitions" (2018), WHO TRS 1019 Annex 5 (2019), and PIC/S PI 041-1 align here.
Both. Retrospective validation is a common entry point. The agent reads the legacy documentation (URS, change control history, prior IQ/OQ/PQ packages, deviation logs), reconstructs the trace graph, identifies gaps against current GAMP 5 expectations, and drafts the remediation evidence pack.
For forward-looking projects the agent runs the full lifecycle from URS to PQ. For legacy systems it reconciles the past and brings it under continuous review going forward.
Bi-directional, validated connectors. Three usage patterns:
Pattern A - Author here, archive there. Validate·AI authors and executes; the existing system of record receives the signed, validated artefact and stores it as the system-of-record copy. Most common pattern with Kneat and ValGenesis customers.
Pattern B - Author there, augment here. The existing system stays the authoring environment; Validate·AI runs deviation triage, RTM maintenance, ALCOA+ audit, and inspection-readiness on top of the records.
Pattern C - Replace the document layer, keep the eQMS. Validate·AI becomes the validation-content layer; deviations, CAPAs, and change controls flow to your eQMS (TrackWise, Veeva Quality, MasterControl, ETQ) as before.
They see three things, on request, in real time:
The narrative. Inspector walkthrough mode generates a guided narrative from your VMP through every system, every deviation, every CAPA - in the language the inspector expects. PMDA-style? FDA-style? MHRA-style? Configurable.
The defence pack. Every action attributable, every record tamper-evident, every decision citable to the SOP clause and the standard. Generated on demand, exported as a regulator-readable package.
The trace graph. Live, queryable, navigable. The inspector can ask "show me the test evidence for URS-014" and get it in two clicks.
Yes, yes, and yes.
The platform is validated as a GAMP 5 Category 4 baseline. Platform validation evidence pack available to qualified prospects under MNDA - includes Validation Plan, Risk Assessment, IQ/OQ summary, Traceability Matrix, and Configuration Management Plan. Site-specific configuration is validated as part of the customer's lifecycle.
The AI model layer is governed under a separate model-governance regime aligned to EU AI Act Articles 12-13, NIST AI RMF, and ICH Q9(R1) risk-based AI lifecycle. Model cards per agent. Change control via signed releases. Drift detection and post-market monitoring engineered in.
All in scope.
Method validation per ICH Q2(R2) (2023). Specificity, linearity, accuracy, precision, range, detection limit, quantitation limit, robustness. Method transfer per USP<1224> with comparative testing, co-validation, or revalidation as appropriate.
Cleaning validation with health-based exposure limits (PDE/ADE) per EMA Q&A on Implementation of Risk-based Prevention of Cross-Contamination, toxicology-driven worst-case limits, and validated cleaning procedures.
Process validation per the FDA 2011 Guidance and EMA equivalent. Stage 1 (process design), Stage 2 (process performance qualification, PPQ), Stage 3 (continued process verification, CPV). ICH Q8/Q9/Q10 aligned.
Yes. ATMP and cell-and-gene have unique validation pressure - personalised batches, vein-to-vein traceability, chain of identity, very tight aseptic envelopes. The platform handles per-batch evidence assembly in line with manufacturing, chain-of-identity and chain-of-custody captured as data not paper, and Annex 1 (2022) aseptic-process alignment.
Annex 1 (2022) contamination control strategy (CCS), pre-use post-sterilisation integrity testing (PUPSIT), media fills, environmental monitoring - all in scope. The agent maintains the CCS as a living artefact across sites, products, and inspection cycles.
The model layer is governed under a documented regime aligned to multiple frameworks:
EU AI Act - high-risk AI system documentation per Articles 12 (record-keeping) and 13 (transparency). Risk management per Article 9. Human oversight per Article 14. Post-market monitoring per Article 72.
NIST AI Risk Management Framework - GOVERN, MAP, MEASURE, MANAGE functions instantiated for each agent.
ICH Q9(R1) - risk-based quality risk management applied to the AI lifecycle. Model cards per agent, signed releases, drift detection, performance monitoring on production data.
ISPE GAMP Good Practice Guide on AI/ML in GxP - qualification envelope for the model layer itself.
We sit above your validation stack. We do not replace it.
Native connectors for the systems of record your QA already runs. Validation, eQMS, MES/EBR, historians, CMMS, identity.
Kneat Gx
Bi-directional connector for validation lifecycle and signed records.
ValGenesis VLMS
Round-trip for validation lifecycle data and approval signatures.
Veeva Vault Validation Mgmt
Native handoff into the Vault Validation tenant.
Veeva Vault Quality
QualityDocs, QMS, Training - agent-authored content live.
MasterControl
Documents, processes, training events synced with agent actions.
Sparta TrackWise Digital
Deviation, CAPA, change control via validated connector.
ETQ Reliance
Nonconformance, audit, supplier management workflows.
Werum PAS-X
Manufacturing execution and batch record data for PV and PPQ.
Rockwell PharmaSuite
EBR, recipe authoring, operations review handoff.
AVEVA PI / OSIsoft PI
CPV feed - process data into PQ and continued verification.
Maximo · SAP PM
Equipment master data, maintenance, calibration records.
Okta · Entra ID
SSO, SCIM provisioning, RBAC enforced at retrieval.
We do not replace your validation system of record. We sit above it - and feed it cleaner records than it has ever seen.
The inspector walks in. You walk them through.
Three things, on demand, in real time - because inspectors do not wait.
Always-on defence pack
Every action attributable. Every record tamper-evident. Every decision citable to the SOP clause and the standard. The defence pack is the platform's resting state - not a quarterly exercise.
Inspector walkthrough mode
When the regulator arrives, generate a guided narrative from your VMP through every system, every deviation, every CAPA - in the language and structure the inspecting authority expects.
Audit trail review at scale
The sample sizes regulators expect (100 percent) at the throughput humans can sustain (1-2 percent). The agent reviews every record, every cycle. Findings escalated; rest archived.