Qualification, then continuous requalification.
Six stages, end-to-end. The agent runs every one against your engineering package, your EM program, and your site contamination control strategy.
Zone classification
ISO 14644-1 Grade A/B/C/D. Layout mapped to grade boundaries.
Design qualified
IQ drafted from P&ID and equipment data sheets.
Operational qualified
Airflow visualisation, recovery, HEPA integrity.
Performance qualified
At-rest and in-operation particles, viables, EM baseline.
Live EM ingest
Viable and non-viable monitoring trended, alarmed, archived.
Requalification
Triggered by data, not by calendar. Scope sized to the excursion.
Six capabilities, one per stage. Same trace graph.
HVAC, WFI, PW, Pure Steam, CDA, and zone classification - one agent across the qualification lifecycle, then across the continuous one.
Zone classification to ISO 14644-1.
The agent maps the physical cleanroom layout to grade boundaries: Grade A (Class 5 in operation), Grade B (Class 5 at-rest / Class 7 in operation), Grade C (Class 7 / Class 8), Grade D (Class 8 / unclassified). Door interlocks, pass-through hatches, and airlock cascades are captured as objects on the graph - not as PDFs.
The output is a cleanroom certification matrix: per zone, per state (at-rest / in operation), per classification, with the qualifying particle counts and viables linked. The matrix is the live representation; the certificate is the export.
HVAC IQ from your P&IDs.
The agent reads the engineering package - P&IDs, equipment data sheets, AHU schedules - and drafts the IQ protocol against the as-built. Filter integrity, airflow direction and velocity, pressure cascade, temperature and RH set-points, and recovery time all enter the protocol as testable acceptance criteria.
OQ adds dynamic conditions: airflow visualisation (smoke study) acceptance per Annex 1, recovery time per ISO 14644-3, and HEPA integrity per ISO 14644-3 or ASTM. The execution records are graph-linked to the engineering package; a P&ID change surfaces the qualified state for review.
WFI, PW, Pure Steam. Sanitised, sampled, signed.
Water systems are qualified through their sanitisation cycle, their conductivity and TOC profile, their endotoxin limits (where applicable), and their microbial counts. The agent integrates the water-monitoring historian, ingests the continuous sample stream, and assembles the qualification report against EU Pharmacopoeia and USP <1231>.
Pure Steam qualification adds non-condensable gases, dryness fraction, and superheat. The agent runs both as the same workflow - different acceptance criteria, same trace graph - and routes excursions to QA with a scoped impact assessment.
Environmental monitoring, live.
Viable monitoring (settle plates, contact plates, active air) and non-viable monitoring (particle counts) feed the EM trending engine. Alert and action limits are derived from baseline data - per zone, per shift, per activity - rather than from generic thresholds.
Trends are visible per zone, per shift, with isolate identification linked where the lab returns the result. The MHRA, FDA, and EMA expectation that EM data is reviewed continuously - not at the year-end audit - is the resting state, not an exercise.
Requalification by data, not by calendar.
Alert and action-limit excursions surface requalification triggers in real time. The agent scopes the minimum requalification - which test, which zone, which condition - rather than re-running the whole PQ. A pressure-cascade excursion does not pull HEPA integrity back into scope unless the trace graph says it must.
Where the customer is on a calendar-based requalification (legacy programs often are), the platform overlays the trigger-based view alongside it - the calendar becomes the upper bound, not the operating cadence.
Per utility, per zone, per site.
One trace graph per utility, per zone, per site. The Cork site's PMDA defence pack does not drag the Basel site into scope. When the inspector asks for the HVAC PQ on Cleanroom B at the Tokyo facility, the platform serves it - current revision, signed, with the EM trend chart that justifies the requalification interval.
Site-level deviation patterns are visible to corporate QA without flattening the per-site validated state. The global view exists for trending; the local view stays defensible.
The controls the platform operationalises.
Annex 1 (2022) is a system of interlocking controls, not a checklist. The platform makes each control a live object on the trace graph - not a paragraph in a binder.
- Contamination Control Strategy (CCS)A living artefact across sites, products, and inspection cycles. Updated on every change.
- Quality Risk Management for sterileICH Q9(R1) applied to the aseptic envelope - per process, per product, per zone.
- Personnel qualification linkageOperator training records cross-referenced to the procedures, the zones, and the gowning programme.
- Environmental monitoring programmeEM plan as data: sample points, frequencies, acceptance limits, trend rules, isolate identification.
- Aseptic Process Simulation (media fill)Records assembled per batch; intervention map maintained; deviations triaged against the simulation.
- Premises and equipment qualification linkageCleanroom qualified state cross-references the cleaning lifecycle and the equipment qualified state.
The standards we operate against. Cited, not implied.
Annex 1 (2022)
Manufacture of sterile medicinal products. Contamination control strategy as the operating frame.
ISO 14644-1 / 14644-2
Classification of air cleanliness and monitoring to demonstrate compliance over time.
EudraLex Annex 15
Qualification & Validation. The Q in HVAC/Water/Pure Steam.
USP <1116>
Microbiological Control and Monitoring of Aseptic Processing Environments.
WHO TRS 961
Where utilities support cold storage - cold-chain qualification linkage.
Part 11
Electronic signatures on every requalification record and signed report.
Five questions a Head of Sterile Operations will ask.
Smoke studies are captured as video evidence with the operating condition (at-rest / in operation), the zone, the date, the operator, and the acceptance review explicitly recorded. The agent indexes the video against the intervention map so that “show me the smoke study covering the vial-loader feed” is one query, not a four-hour archaeology session.
Where Annex 1 requires re-demonstration after a process change, the trigger fires on the trace graph - the change object surfaces the affected smoke studies and routes the new study to scheduling.
EM data sourced from a system that is not itself qualified is flagged at ingestion. The data is still ingested for trending and pattern recognition, but acceptance decisions cannot be drawn from it - the agent enforces the boundary explicitly. A separate workstream proposes the validation route for the source system: qualification scope, gap analysis, remediation plan.
Most life-sciences customers run a mix - qualified Vaisala for particle counts, semi-qualified spreadsheet plates, ad-hoc swabs. The platform models the integrity of each source as a property of the data, not as a global assumption.
Scope is derived from the trigger and the change-impact graph. An action-limit excursion on non-viable particles in Grade B may scope to a particle-count requalification on that zone only, with a contained CAPA. A pressure-cascade failure between airlock and Grade A pulls airflow visualisation back into scope, but does not pull HEPA integrity unless the integrity test interval has also lapsed.
Full requalification is reserved for changes that touch the qualified state itself - AHU replacement, ducting modification, HEPA bank change. The agent shows the rationale on the trace graph so the QA decision is auditable, not opaque.
Yes. Media fill is the process-level demonstration that the cleanroom-and-equipment system holds the aseptic envelope. The agent assembles the media fill record per batch, maintains the intervention map (planned and unplanned), reviews the simulation against Annex 1 acceptance criteria (zero positive units in the standard case), and routes contamination findings to deviation triage.
Where a media fill fails, the trace graph propagates impact: which products were released after the previous successful simulation, what their disposition is, and what re-simulation scope applies. The clean-up is bounded by data, not by precaution.
One graph, two views. Validate·AI owns the qualified state of the cleanroom and the cleaning validation; Operate·AI owns the execution of cleaning events on the floor. Both write to the same underlying objects: the equipment qualified state, the cleaning validation in force, the execution record, the deviation.
When a cleanroom requalification updates the qualified state, the Operate·AI scheduling layer reads the new condition automatically. When an Operate·AI cleaning execution finds an unexpected residue, the trace surfaces it back into Validate·AI as a candidate trigger for the cleaning validation. The two products share a system of truth, not a connector.