How it works

QP reasoning logic,
not a chatbot.

ChatQP is not a general chatbot. It first works out what kind of quality problem it is looking at. Then it pulls the right sources, applies hard GMP rules outside the model, and gives the QP a structured output to review.

A disciplined method, not a chat window.

ChatQP is not a general chatbot. It works out the type of quality problem in front of it, then reviews the case across the relevant scientific and regulatory domains, retrieving the strongest regulatory sources first. Hard GMP rules run deterministically in code, outside the model, and cannot be overridden by a risk-based argument. From there it returns a structured, source-cited position for the Qualified Person to review — with the regulatory basis, the key risks, and the open questions set out. It never certifies a batch: it prepares the decision so a person can make it and defend it.

Four positions. Clear every time.

ChatQP never certifies a batch. It states one of four positions for the QP to review — wording that says whether certification is supported, never whether the batch is released.

CERTIFICATION SUPPORTED

All requirements are met. The QP can certify, with the regulatory basis stated.

SUPPORTED WITH CONDITIONS

Supportable once a documented, bounded justification and the listed actions are in place.

NOT CURRENTLY SUPPORTED

An open concern blocks certification until it is investigated. A specific path forward is given.

HARD BLOCK

A hard GMP rule was triggered. No risk-based argument overrides it.

The QP owns the decision.

ChatQP does not certify batches. It lays out the evidence and flags the risks. The Qualified Person stays responsible for the final decision under EU GMP Annex 16. That is not just legal caution. It is the point of the whole design.

Design commitment

Hard rules sit outside the model. Sources are cited. The reasoning trace is visible. Nothing leaves your network. The QP reviews, challenges, and decides.

See benchmark outputs Annex 22 alignment