Validated AI for trials, safety and quality.
The life sciences pack brings safety cases, trials and quality records to Nexus. AI reads case reports, drafts narratives and spots deviations, with every source shown. Qualified people decide. Every record is sealed, with the audit trail inspectors expect.
| Case | Seriousness | Due |
|---|---|---|
| ICSR-26-0912 HB-207 | Serious | Due in 13 days |
| ICSR-26-0908 HB-114 | Non-serious | Due in 84 days |
| ICSR-26-0905 HB-207 | Possible duplicate | — |
Patient, 64, admitted to hospital with severe low blood pressure two days after the first dose of HB-207 1. Already on two blood pressure medicines 2.
The AI drafts. A medical reviewer assesses every serious case.
Less transcribing. More judgement.
Safety, clinical and quality teams spend much of their day moving facts from one document to another. The pack does that work and shows its sources, so experts spend their time deciding.
Safety cases processed with sources
Reports from email, calls and literature are read, coded and checked for duplicates. Every fact links back to where it came from.
Deviations caught early
Trial and batch records are checked as they arrive. Possible serious breaches are flagged with the seven-day clock already running.
Validation you can hand to an inspector
Each AI use keeps its intended use, test results and limits, alongside an audit trail that cannot be quietly changed.
One record. A lens for each role.
Each role sees its part of the same records, with an audit trail on everything.
- Case processor
New safety reports, extracted facts, duplicate checks and narratives to finish.
- Medical reviewer
Serious cases, causality to assess and the evidence behind each fact.
- Clinical operations
Deviations by site, possible serious breaches and their deadlines.
- Quality assurance and Qualified Person
Batch records, findings and what stands between a batch and release.
- Validation lead
Each AI use, its tests, its limits and every change made to it.
From first report to a sealed assessment.
Four steps, the same for every case. Follow one safety report step by step →
Receive
Reports arrive by email, phone, partner feeds and literature.
Read
AI extracts the facts, codes the terms and looks for duplicates, with sources.
Assess
A medical reviewer judges seriousness and causality.
Seal
The assessment, the narrative and the submission are sealed.
The screens your teams use.
Sample data from a fictional company, Halden Bio.
Every case, with its deadline
Seriousness and the reporting deadline for every case, and a summary that shows its sources.
| Case | Seriousness | Report due |
|---|---|---|
| ICSR-26-0912 HB-207 | Serious | Due in 13 days |
| ICSR-26-0908 HB-114 | Non-serious | Due in 84 days |
| ICSR-26-0905 HB-207 | Possible duplicate | — |
Patient, 64, admitted to hospital with severe low blood pressure two days after the first dose of HB-207 1. Already on two blood pressure medicines 2.
The AI drafts. A medical reviewer assesses every serious case.
Deviations flagged by site
Deviations are read from site files and monitoring notes. Possible serious breaches are flagged at once.
| Site | What happened | Could be serious? |
|---|---|---|
| Site 12, Brno | Dose given outside the window | Check |
| Site 04, Cork | Consent form version out of date | Likely |
Two patients signed version 3 of the consent form after version 4 was approved 1. Version 4 added a new risk 2.
If it is a serious breach, the sponsor must notify within seven days. Clinical operations decides.
Batch records checked before QA
Every step, limit and calibration is checked. QA reviews the findings; the Qualified Person releases.
| Check | Result |
|---|---|
| All steps signed | Yes |
| In-process limits | Within limits |
| Equipment calibration | Mixer M3 overdue 2 days |
| Deviations closed | Yes |
Release is always by the Qualified Person.
A technical file for each AI use
Intended use, test data, results and limits, kept with the system and versioned.
Extract facts from safety reports for a case processor to check. Never decides seriousness or causality.
- Languages
- EN, DE, FR, ES, IT
- Not for
- Handwritten forms
- Human check
- Every case
| Test set | Result | Version |
|---|---|---|
| Held-out reports, 600 | Passed | v7 |
| Duplicates, 120 pairs | Passed | v7 |
| Change: new model | Re-test due | v8 |
Any change goes through change control and re-testing before use.
Every assessment, with its audit trail.
When a reviewer signs off a case or QA closes a finding, the pack seals the record with everything it relied on.
The source documents, the facts extracted, every edit, the assessment and who signed it.
Inspectors and auditors can check the seal with the open verifier.
An audit trail that cannot be quietly changed is what computerised-system rules ask for.
Try a seal check yourself on the home page.
GxP and safety rules, mapped to the evidence.
The pack links each obligation to the records that answer it, with the audit trail inspectors look for.
- EU GMP Annex 11EudraLex Volume 4Computerised systems: validation, audit trails and e-signatures
- EU GMP Annex 22, draftConsultation draft, July 2025AI in GMP: intended use, test data and explainability
- Clinical Trials RegulationRegulation (EU) No 536/2014Trial conduct, transparency and serious breach notice
- ICH E6(R3)Good clinical practiceRisk-based quality and data integrity, in effect from 23 July 2025
- PharmacovigilanceImplementing Regulation (EU) No 520/2012Pharmacovigilance systems and their quality
- MDR and IVDRRegulations (EU) 2017/745 and 2017/746Software that is a medical or diagnostic device
- EU AI ActRegulation (EU) 2024/1689 · Annex IHigh-risk rules for AI in medical devices from 2 August 2028
- GDPRRegulation (EU) 2016/679 · Art. 9Health data handled with a legal basis and safeguards
AI Act dates as amended by Regulation (EU) 2026/1744, the Digital Omnibus, published in the Official Journal on 24 July 2026. Not legal advice.
Eight modules, one audit trail.
Turn on the modules each function needs. All of them share the same validated core and sealed records.
01Case intake
Reads safety reports from email, calls, partner feeds and literature, and checks the minimum criteria.
- Goes in
- Reports, attachments
- Comes out
- A draft case with facts and sources
- Signs off
- Case processor
02Coding and duplicates
Codes events and products and looks for duplicate reports.
- Goes in
- Draft cases
- Comes out
- Coded cases, possible duplicates
- Signs off
- Case processor
03Narratives
Drafts the case narrative from the facts, with a source for every sentence.
- Goes in
- Coded case
- Comes out
- A narrative ready for review
- Signs off
- Medical reviewer
04Signal detection
Looks for patterns across cases and literature and explains each one.
- Goes in
- Case database, literature
- Comes out
- Possible signals with evidence
- Signs off
- Safety physician
05Trial deviations
Reads site files and monitoring notes and flags possible serious breaches.
- Goes in
- Site files, monitoring reports
- Comes out
- Deviations with deadlines
- Signs off
- Clinical operations
06Batch record review
Checks electronic batch records before QA review.
- Goes in
- Batch records, calibration logs
- Comes out
- Findings for QA
- Signs off
- Qualified Person
07Validation pack
Keeps intended use, tests, results and limits for each AI use, versioned.
- Goes in
- Test data, change requests
- Comes out
- A validation file ready for inspection
- Signs off
- Validation lead
08Inspection room
Gathers sealed records for an inspector, linked to the rule they answer.
- Goes in
- Sealed records
- Comes out
- An evidence set an inspector can verify
- Signs off
- Head of quality
Validated once, run where you need it.
Run the pack in our EU cloud or in your own data centre. Sites with strict controls can run offline. Where Nexus runs →
Our EU cloud
The company in its own private space. Data stays in Europe.
Your data centre
Inside your validated environment.
Fully offline
For manufacturing sites with no outside link.
What people ask about the life sciences pack.
Does the AI decide causality or seriousness?
No. It extracts facts and drafts. A medical reviewer assesses every serious case and signs the assessment.
How is the AI validated?
Each AI use keeps its intended use, test data, results and limits in a versioned file. Changes go through change control.
Is there an audit trail?
Yes. Every view, edit and decision is recorded, and sealed records cannot be changed without the check failing.
Can it release a batch?
No. It checks the record and lists findings. The Qualified Person releases.
Is patient data used to train shared models?
No. Your data is not used to train shared models. Models run in the place you choose.
Can inspectors check the records?
Yes. Sealed records can be checked with the open verifier, without an account.
Other packs
See it on one of your own cases.
Bring a safety case or a deviation that took too long. We will show you how it runs on Nexus.