MIRA · Governed AI AssistanceAdvisory + Human-in-the-Loop

AI assistance that stays inside the workflow.

MIRA supports defined tasks inside Verixa’s GMP Core — Document Control, Deviation, RCA, CAPA and Inspection Readiness. It can surface missing information, organise cited source material and draft workflow content for review. MIRA does not approve, close, sign, release or dispose of regulated records. A qualified human reviews AI-assisted content before it is accepted into a regulated record — and the record is designed to show both the assistance and the decision.

MIRA whitepaper is gated — requested via the team.

What MIRA Does

An assistant for your QA team — not a decision-maker

MIRA supports the work of surfacing missing context, organising source material and drafting content for review. The judgment stays with your qualified reviewers, and the decision evidence is designed to be logged according to the workflow’s implemented and verified control scope.

Surfaces gaps

Reads your GMP records and surfaces gaps, inconsistencies, and missing evidence for a human to review.

Drafts deviation summaries

Drafts deviation summaries and narratives that your QA team edits, accepts, or rejects.

Suggests CAPA actions

Suggests candidate CAPA actions and root-cause angles — as labelled suggestions, never as decisions. MIRA does not draft the CAPA record itself: that authorship, and the decision, stay human.

Provides source references

MIRA is designed to provide relevant source references with its suggestions for reviewer assessment.

Records the system interaction

The control design records the applicable model identifier, configuration version, source context and generated output.

Records the reviewer's disposition

The reviewer's accept / reject / modify action, and the e-signature where applicable, is maintained with the workflow record.

MIRA is not the product by itself. The value comes from the controlled workflow, human-review gates and evidence context around its use.

What MIRA Does Not Do

The hard lines around GxP-critical decisions

MIRA is never the sole path for a critical GMP decision. The intended workflow requires qualified human review before AI-assisted content is accepted into a regulated record.

These refusals are not limitations. They are the governance evidence — each one is a boundary you can show an inspector, on the record.

MIRA does not approve, close, sign, dispose, or release any GxP record.

MIRA does not write to a GxP field without your explicit accept.

MIRA does not set deviation severity. It may suggest a severity classification as labelled advisory input — but a qualified human sets it, and a second human reviews the escalation.

MIRA does not draft the CAPA record — ask it to, and it declines.

MIRA does not make critical GMP decisions — batch release, OOS classification, CAPA disposition.

See it in practice

One deviation — MIRA assists, then declines

In the end-to-end GMP scenario, MIRA proposes investigation themes on a critical deviation — and the record preserves its suggestion and the human conclusion side by side, each attributed. Then someone asks it to draft the CAPA, and it declines.

1

Critical deviation raised

2

MIRA proposes RCA themes

advisory

3

A different human concludes

attributed

4

Asked to draft the CAPA

MIRA declines

Governance Architecture

The control documents behind MIRA

MIRA is governed by a documented set of controls, available to founding partners as a gated download. Each defines how the model is used, assured, and changed.

Request the MIRA architecture pack
The control set5 documents · gated
  • AI Governance V7 Plan
  • Provider Assurance Pack
  • Intended Use Statement
  • Acceptance Criteria
  • Change Control rules
Regulatory Framing

Designed against the standards taking shape now

EU GMP Annex 22 (Draft 2025)

Treated as an internal-control reference — generative AI prohibited from sole-path critical GMP decisions.

EU AI Act (Reg. 2024/1689)

Role-specific, obligation-specific. Phasing in 2026–2027. Verixa makes no claim of EU AI Act compliance.

FDA CSA Final Guidance (Feb 2026)

Risk-based, critical-thinking approach to computer software assurance informs our design.

GMLP — Good Machine Learning Practices

Good Machine Learning Practice principles guide model lifecycle and oversight.

21 CFR Part 11 audit-trail extension

AI inferences and human decisions extend the electronic audit trail as first-class records.

Compliance is a customer determination, not a vendor claim. Verixa provides the architecture and evidence; the customer validates intended use under its own quality system.

Which workflow should Verixa map first?

Select one current quality process. We will show how Verixa could support it, where AI assists, where human decisions remain mandatory, and what validation responsibilities stay with your organization.