For healthtechs, NHS trusts, pharma, MedTech, and clinical research

Clinical AI
the NHS can sign off.

Production NLP for clinical documents, trial-matching, prior-authorisation, and structured patient records — mapped to SNOMED CT / ICD-10, NHS IG-aware, MHRA-ready when the system is a medical device.

Already running a 10x patient-record-processing engagement
NHS IG · MHRA · EMA · CE-marked-as-medical-device aware
SNOMED CT & ICD-10 ontology mappings included

Patients are waiting on data your clinicians cannot process fast enough.

Most healthtech and clinical-ops teams we work with have the same three problems — and adding analysts has stopped being a viable answer.

01

Unstructured clinical notes

GP correspondence, discharge summaries, scanned referrals — the data is in there, but it is in 12 formats, 4 languages, and bound by NHS Information Governance rules.

02

Trial matching at scale

Eligibility checks done by hand against 200+ medical entities, with a delay measured in days. Patients miss their trial windows because of operational latency, not clinical eligibility.

03

Regulator-ready AI

MHRA Software-as-a-Medical-Device, the EU AI Act's high-risk classification for health, NHS DSP toolkit, and the AI Safety Institute — all stacking into a single sign-off your second-line cannot wave through.

Clinical AI built to pass second-line review.

We ship one tightly-scoped clinical AI engagement at a time, in 6 to 10 weeks, with documentation built for MHRA / NHS DSPT / EU AI Act audits.

  • Biomedical document NLP Named-entity recognition trained on UK / EU clinical text. Extract medications, diagnoses, dosages, procedures — with confidence bands and a human-in-the-loop escalation lane.
  • Standards-mapped output Everything maps to SNOMED CT and ICD-10 (and to NHS GP Connect / FHIR where the integration calls for it). Interoperable by construction, not after the fact.
  • Trial-matching & prior-auth Patient profiles checked against trial eligibility criteria or insurer criteria in seconds, with a cited audit trail your clinical lead can defend.
  • MHRA / AI Act documentation pack Intended purpose, risk assessment, clinical evaluation, post-market surveillance plan, and EU AI Act Annex IV docs — produced in your template, ready for notified body review.
Stack we route across
ModelsAnthropic Claude · Mistral · open-weight biomedical models
OntologiesSNOMED CT · ICD-10 · LOINC · RxNorm
HostingEU-resident: Scaleway · OVHcloud · NHS-approved AWS regions
StandardsHL7 FHIR · NHS GP Connect · DICOM (where relevant)
Compliance postureNHS DSPT · MHRA SaMD · EU AI Act high-risk · UK ICO
Audit & evaluationCited sources, retrieval traces, ongoing clinical eval

The clinical AI stack, live.

A representative view of the clinical NLP & trial-matching engine we ship for healthtechs and NHS trusts. Click between tabs, switch languages — everything in here is built to work in your clinicians’ and patients’ language.

Insightrix · Clinical AI Console Live Clinical NLP Demo
Aaru.eu

Clinical NLP · discharge summary

Free-text clinical document → structured entities → ontology codes. Confidence bands every step, human-in-the-loop escalation lane.
live EN-GB
Notes processed today
3,184
+12%
Sensitivity (eval)
92%
+1pp
Avg processing
218ms
-22ms
Escalated to clinician
8%
-2pp
GP discharge summary · doc_2418.pdf
Patient presented with persistent hypertension. Initiated amlodipine 5mg OD. Follow-up in 4 weeks. Existing diagnosis: Type 2 diabetes mellitus, well-controlled on metformin 1000mg BD.
diagnosis medication follow-up
AI-extracted entities · confidence
Hypertension · diagnosis0.97
T2DM · diagnosis0.96
Amlodipine 5mg OD · medication0.94
Metformin 1000mg BD · medication0.93
4 weeks · follow-up window0.91
SNOMED CTICD-10RxNorm

Ontology mapping · SNOMED CT & ICD-10

Free-text entities mapped to standards by construction. Interoperable with NHS GP Connect, HL7 FHIR R4 and DICOM.
SNOMED 2026-04 FHIR R4
EntitySNOMED CTICD-10RxNormConf.Status
Hypertension38341003I100.97mapped
Type 2 diabetes mellitus44054006E110.96mapped
Amlodipine 5mg3875620001973610.94mapped
Metformin 1000mg3725670098610070.93mapped
Follow-up consultation350250070.91mapped
eGFR802740010.88review
Coverage by domain (eval suite)
Diagnoses96%
Medications93%
Procedures89%
Lab values84%
Interop targets
HL7 FHIR R4ready
NHS GP Connectready
DICOM · imagingready
OMOP CDMoptional

Trial matching · pat_4421 vs NCT01234567

Patient profile checked against eligibility criteria in seconds, with a cited audit trail your clinical lead can defend.
pat_4421 match 0.94
Patient profile · pat_4421
Age62
SexF
DiagnosesT2DM · HTN
HbA1c7.4%
eGFR82
BMI28.4
Current medsamlodipine 5mg OD · metformin 1000mg BD
Eligibility · NCT01234567
Age 50–70pass
T2DM > 3 yearspass
HbA1c 6.5–9.0%pass
eGFR > 60pass
No CKD stage ≥ 3pass
Stable insulin regimen 90dverify
Match score
0.94
5/6 criteria met. Insulin regimen needs verification before referral.
Recommended actionSchedule eligibility verification with patient pat_4421 in next clinic slot. Coordinator notified. Match probability post-verification: 0.97.

SaMD documentation · clinical NLP v1.4

Software-as-a-Medical-Device. MHRA Class IIa. Documentation pack ships in every clinical engagement — your notified body review starts here.
MHRA Class IIa v1.4.2
High-risk AI under EU AI Act · Annex IIIHealth-domain AI is high-risk by classification. Clinical evaluation, post-market surveillance, and adverse-outcome logging are mandatory before placing on market.
FieldValue
Intended purposeDecision support · eligibility check · not autonomous diagnosis
Intended usersTrial coordinators · clinical research nurses · GP / consultant oversight
MHRA classIIa
EU AI ActHigh-risk (Annex III · health)
Clinical evaluationN=2,400 · sensitivity 92% · specificity 89%
Human oversightClinician sign-off required before referral. AI flags only, never auto-refers.
Post-market surveillanceDrift alerts · adverse-outcome log · quarterly clinical eval refresh
Training dataUK clinical corpus 42k notes · SNOMED CT 2026-04 · bias audit Q3 2025
Latest clinical eval
0.92
SensitivityN=2,400 · UK NHS corpus
Specificity89%
Drift score0.3%
Adverse outcomes (30d)0
14:42:18EVALQuarterly clinical eval passed · sensitivity 0.92 · specificity 0.89
14:41:54DEPLOYClinical NLP v1.4.2 promoted · bias audit annexed
14:40:32PMSPost-market surveillance log appended · 0 adverse outcomes (30d)
14:38:18REVIEWMHRA Annex IV pack shared with notified body · @raj
NHS DSPT · MHRA Class IIaSNOMED CT 2026-04 · ICD-10 WHO 2024extraction sensitivity 92%aws-eu-london · NHS-approved region live req 142,481

What our engagements typically deliver.

10× patient-record processing throughput
92% extraction accuracy on UK clinical text
6–10 weeks kick-off to production

Tell us about your clinical workflow.

A 30-minute call. We will look at your actual workflow, the regulator you answer to (MHRA / EMA / NHS / ICO), and tell you which clinical AI use case to ship first — and which to defer until the regulatory shape is clearer.

  • We will tell you if a use case crosses the SaMD line and needs MHRA registration first.
  • NDAs available before the call. DPA-ready engagement template.
  • You'll leave with a one-page scoped roadmap, even if you do not engage.
UK practice lead Raj is at raj.singh@insightrix.co.uk.

By submitting, you agree to our privacy policy. We respond within 24 working hours and never share your details. NHS / DPA-ready.

Common questions before we talk.

Are you a regulated medical-device manufacturer?

We are not. We deliver clinical AI systems for clients who are (or who need to register as one). Our role is to build, document, and harden the system so your MHRA / EMA submission is straightforward; the placing-on-market and post-market obligations sit with you as the manufacturer.

Can you work inside our NHS trust's DSPT and IG framework?

Yes. We have worked under NHS Information Governance, signed DSPT-aligned DPAs, and deployed on NHS-friendly cloud regions. Data residency stays where you need it.

How do you handle clinical evaluation?

We do not do clinical evaluation ourselves — that's the manufacturer's clinical-affairs function. We do produce the engineering artefacts the clinical evaluation needs: ground-truth datasets, retrieval / extraction accuracy reports, drift monitoring, and adverse-outcome logging.

How do you scope and engage?

Every engagement is scoped after a 30-minute discovery call and a clinical-workflow walk-through. We send a written, fixed-scope proposal — with timeline, clinical and regulatory deliverables, and a transparent quote — within 5 working days.

Ship clinical AI your second line will sign off on.

Book a 30-minute call. We will scope a use case that delivers in weeks, with the audit trail and regulator-ready documentation built in from day one.

Book a Free Call