⚠ Methods & limitations notice — read before use
NexaPV Assist · development build — not a validated system
1. What actually produces the output you see — not just "AI"

It's easy to read a notice titled "AI use" and assume everything here comes from a language model. It doesn't. Most of what this tool produces comes from three other sources, with AI as one optional layer on top:

  • Deterministic local logic — keyword/pattern scans, PRR/ROR/χ² calculations, date and timeline arithmetic, and the Cross-Reporting decision engine all run entirely in your browser with no model involved, AI key or not.
  • Your own uploaded reference material — company templates, prior RMPs/PSURs, SDEA matrices, and reference safety information you upload are read directly and used as the source structure and content for drafts. Several of the built-in section templates and structures in this tool were themselves authored by Dr. Syed Mudasir Ahmad from real regulatory documents and precedent, not generated by a model.
  • Web/search-based lookups — openFDA, PubMed/MEDLINE, and (where a key is set) AI-directed web search against health-authority sites retrieve real, citable public pages. These are searches of existing public sources, not something invented.
  • AI drafting (Anthropic's Claude, optional) — only where you've added your own API key, and only for narrative prose, content review, and web-search synthesis specifically — never for the deterministic calculations or your uploaded source content, which are handled by the methods above regardless of whether a key is set.

Whichever combination produced a given output, every output is a draft proposal for a qualified person to review, correct, and approve — never a finished assessment, decision, or regulatory record. Nothing produced here is fit to be filed, submitted, or entered into a safety database without independent verification against the source data by an appropriately qualified reviewer.

Where AI drafting specifically is used, this reflects the human-centric, risk-based approach set out in the EMA's Reflection paper on the use of Artificial Intelligence in the medicinal product lifecycle (adopted September 2024), which expressly covers post-authorisation uses including adverse event report management and signal detection, and the risk-based credibility assessment framework in the FDA's draft guidance Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products (January 2025), together with the joint FDA–EMA Guiding Principles of Good AI Practice in Drug Development (January 2026).

2. Validation status — please read carefully

This is an unvalidated development build. Specifically, none of the following has been done:

  • No computerised system validation (no IQ/OQ/PQ, no GAMP 5 lifecycle documentation).
  • No credibility assessment for any defined context of use, as the FDA framework requires — model influence and decision consequence have not been formally assessed.
  • Not compliant with 21 CFR Part 11 / Annex 11. The audit trail here is a convenience log; it is not tamper-evident, and there are no electronic signatures.
  • No qualification under any GxP quality system; no change control; no periodic review; no vendor qualification of the underlying AI model.
  • No formal performance monitoring, bias assessment, or drift monitoring of model outputs.

If you intend to use this for work supporting regulatory decisions, it must first be assessed, validated, and approved under your organisation's quality system and IT/InfoSec governance. Regulators may hold AI systems in this space to standards stricter than general industry practice.

3. Known limitations you must account for
  • The AI can be confidently wrong. Large language models can fabricate plausible-sounding facts, citations, dates, and reasoning. Unlike a human error, a fabrication may arrive with fluent supporting justification.
  • MedDRA terms are suggestions only and must be verified against the licensed MedDRA browser for the applicable version. This tool holds no MedDRA licence or dictionary. Numeric MedDRA codes are only present in E2B output if you entered them yourself.
  • Causality, seriousness, expectedness/listedness, and benefit-risk conclusions require a qualified medical assessor. Any value pre-filled here is a starting suggestion under WHO-UMC criteria, not an assessment.
  • Disproportionality statistics (PRR/ROR/χ²) are screening aids only — they reflect only the data you upload, and per GVP Module IX do not validate a signal without qualitative medical review.
  • Text extraction and OCR can misread or silently drop content, particularly in scanned documents and dense tables. Always reconcile against the source document.
  • E2B(R3) output is a structural draft, not a certified submission-ready file, and must be validated against your own database's import specification in a sandbox before any live use.
  • The tool does not learn or improve from use. Errors persist until the code is changed.
4. Accountability remains with you

Use of AI does not transfer, dilute, or share regulatory accountability. The QPPV, MAH, and the qualified individuals performing each activity remain fully responsible for the accuracy, completeness, and compliance of every record and submission — exactly as if no AI had been involved. "The AI produced it" is not a defensible position before any competent authority.

5. Data protection & confidentiality

Where AI features are used, the text you submit is transmitted to Anthropic's API over the internet. Do not paste personal data, patient-identifiable information, or confidential material unless your organisation has approved that transfer and any required agreements and data-protection assessments (e.g. DPDP Act, GDPR) are in place. Local scan, OCR, and file parsing run in your browser; saved settings, credentials, and any persisted log live in this browser's local storage only, and are not secure storage.

6. Ownership, copyright & permitted use

© 2026 Dr Syed Mudasir Ahmad (docmuddie@gmail.com). All rights reserved. This software, including its structure, templates, prompts, and generated document formats, is the proprietary work of the author and is protected by copyright.

It is made available to the registered user named at sign-in, for that user's own use only. Copying, distributing, forwarding, sharing, uploading, sublicensing, reselling, publishing, or otherwise making this tool available to any third party — in whole or in part, modified or unmodified — without the prior written permission of the author is prohibited. Unauthorised reproduction or distribution constitutes copyright infringement and may give rise to civil and/or criminal liability under applicable copyright law. To request permission, or to enquire about a licence, contact docmuddie@gmail.com.

Regulatory references are provided for context and reflect guidance current at the time of writing; they are summaries, not legal advice, and guidance in this area is changing quickly — verify the current text of any document you rely on. This tool is not endorsed by, affiliated with, or certified by the FDA, EMA, CDSCO, ICH, Anthropic, or any regulatory authority.