Industrial software development
Software supporting industrial activities and compliance: calibration, access control, root cause analysis and predictive maintenance.
Case study · Pharmaceutical
AI-assisted Root Cause Analysis with QualityMind
Root Cause AnalysisAI man-in-the-loopDeviations & OOSOn-premiseAnnex 1121 CFR Part 11
The challenge
In the pharmaceutical industry, root cause investigations of deviations and Out of Specification events are often lengthy, depend on the individual investigator's experience and struggle to be exhaustive and traceable. The challenge we set ourselves in developing QualityMind is to introduce AI-based support without handing control of the investigation to the machine and without compromising compliance: data must not leave the site and final validation must remain with the Quality team.
The solution
Event analysis and ranked hypotheses
QualityMind analyzes the deviation or OOS documentation, proposes and ranks root-cause hypotheses and orders their plausibility, giving the investigator a structured working base rather than a blank starting point.
Editable drafts (5-Whys, Ishikawa)
The system generates editable drafts — 5-Whys, Ishikawa diagram, scored report — that the Quality investigator reviews, corrects and completes. They are not final outputs: they are a first draft that speeds up human work.
Man-in-the-loop
QualityMind supports, it doesn't decide: any hypothesis not anchored to evidence is flagged for human review and final validation always remains with the Quality team. It is an assistant to the investigator, not a replacement.
Designed for compliance
The system runs on-premise in an isolated environment, with a 'locked' AI model (no learning in production) and a full audit trail, consistent with Annex 11 and 21 CFR Part 11. The investigation data remains under the site's control.
Frequently asked questions
Does QualityMind decide the root cause instead of the Quality team?
No. QualityMind is a man-in-the-loop tool: it analyzes, proposes and ranks hypotheses and generates editable drafts, but final validation always remains with the Quality investigator. Any hypothesis not anchored to evidence is flagged for human review.
Does the investigation data leave the site?
No. QualityMind runs on-premise in an isolated environment. The AI model is 'locked', meaning it does not learn in production, and every operation is recorded in a full audit trail, consistent with Annex 11 and 21 CFR Part 11 requirements.
What outputs does it produce to support the investigation?
Editable drafts of the main analysis tools: 5-Whys, Ishikawa diagram and a scored report with ranked root-cause hypotheses. They are working bases that the investigator reviews and completes, not final documents.
Do you have a similar challenge?
Let's talk about it. We respond with a concrete proposal, from a team that knows regulated environments from the inside.