Sentinel Capture is the name behind our industry-leading, in-house Intelligent Document Processing platform — designed, built and operated by our own team. It runs on two engines: our own self-learning, in-house engine — trained only on blank forms and sample documents, never on client data — which handles many jobs entirely within our own UK systems; and, where a job needs it, a UK-hosted enterprise AI engine used under strict no-training, zero-retention terms, with your identifiers removed first. It sets us apart from competitors who simply resell the same off-the-shelf capture tools. It's tailored to each client's own document taxonomy, pseudonymises identifying data before it reaches our AI engine, and keeps a full evidential audit trail behind every decision.
Everything described on our AI & Intelligent Document Processing page — the classification, the extraction, the worked medical-records example — runs on Sentinel Capture, the platform our own team designed and built rather than bought in. That means classification is trained around each client's own document taxonomy from the outset, not a generic best-effort approximation.
Owning the platform end to end also lets us build in the things that matter most for records that have to stand up as evidence: pseudonymisation before any call to our AI, a complete audit trail for every classification decision, and a review workflow that puts a person in the loop before anything is finalised.
Identifying details are detected and tokenised before any content reaches our AI engine — full detail in our AI & Data Protection Statement.
Every capture, classification and review decision is logged — built with BS 10008 legal-admissibility principles in mind.
Nothing is finalised on AI confidence alone — every batch passes a review queue, and anything uncertain is flagged before it goes out.
Classification is built around your own document types and sections, not a one-size-fits-all model.
Every user signs in with their own role-based account and multi-factor authentication, with every action attributed and logged.
Handles high-volume batches without a per-page licence — capacity that grows with you rather than your bill.
Documents are scanned and read, page by page.
Two independent detection passes tokenise identifying details in a local vault before anything reaches our AI engine.
Each document is classified and split against your own taxonomy.
A reviewer checks and signs off anything the system flags.
Indexed, audit-logged records land in your system or our portal.
These are screens from Sentinel Capture itself, captured on a synthetic-data demo — no real patient or client records were used to produce these images.
✓ Pseudonymised in our UK facility before any processing — 17 identifiers redacted.
Breakdown: ORGANISATION 3 · CLINICAL DATE 2 · DATE OF BIRTH 2 · NHS NUMBER 3 · ID 1 · PATIENT 1 · PERSON 4 · STAFF 1. The image below — and the copy sent on for content extraction on the right — is the redacted version, not the original; that step never sees a name or a number. Real values live only in an encrypted token vault, access-restricted to named roles and never released. Notice there is no patient or clinician name in the extracted fields — that is deliberate. Identity is handled by a separate, narrower path.
Show original (demo QC view only)Before any content reaches our AI engine, Sentinel Capture runs two independent detection passes and tokenises what they find — names, dates of birth, addresses, NHS numbers — replacing them with local reference tokens.
A reviewer sees the pseudonymised document alongside what our AI extracted, with confidence scoring on every field. Anything below threshold is flagged and held — nothing is finalised without a human sign-off.
Independent NER detection runs before pseudonymisation, not one — reducing what either model misses alone.
Token-to-identity mapping is held in our own environment, never sent to our AI engine.
Low-confidence fields route to a human reviewer automatically — the system doesn't guess and ship.
The same accredited team that stores and scans your records built the platform that reads them.
Designed, built and operated by our own team — not a reseller of someone else's black box.
Identifying data is tokenised before any processing by our AI, by design, not as an afterthought.
Full audit logging built with BS 10008 legal-admissibility principles in mind, from day one.
Runs inside our own accredited UK environment, alongside the rest of our storage and scanning operation.
Send us a representative sample and we'll show you exactly how it classifies, pseudonymises and reviews — then scope it to your volumes and taxonomy.
Talk to us about SentinelAlready a client, or part of our team? Sign in to the Sentinel Capture portal →