OCRXS reads Arabic and English forms — printed and handwritten — into structured fields, flags what it is unsure about, and puts a person in charge of the final value. The source document stays with the record, inside Data X.
Part of Data X by Grid Applications Ltd
What it is
OCRXS is the capture module of Data X. It turns scanned forms into the structured records that FormXS routes and XVS.ai answers from — and it is built around the fact that recognition is sometimes unsure.
Arabic and English, printed and handwritten — including both mixed on one page. Text and named fields are read from the scanned document, not retyped.
On PR-01 the label reads “Quantity / الكمية” and the value is handwritten. Both are read from the same page, without retyping.
Values the recognition is not confident about are marked for review rather than silently accepted. A reviewer confirms or corrects each one beside its place on the page.
The handwritten quantity is read as “72” with low confidence. A reviewer compares it with the page and saves 12 — the original reading stays in the audit history.
The scanned document is retained with the record. Any value can be traced back to the page it came from — during review and in every later audit.
Record PR-2026-00147 keeps the PR-01 scan attached. During Finance review, Omar Al-Rashid can open the exact line the quantity came from.
Demonstration
Run the scan. Fields are extracted with a confidence mark; select any field to see where it came from on the page. One handwritten value is read with low confidence and must be reviewed before the record can pass into the workflow. Change the conditions in the toolbar above to see what raises flags.
Run the scan to extract the fields from the document.
Review queue
Flagged values appear here, one at a time, beside their location on the page.
Recognition quality is not a single number. Clean printed forms read very differently from rushed handwriting on a photocopied page. Accuracy depends on the handwriting itself, the scan quality and the document type — which is precisely why uncertain values are flagged for review instead of being trusted.
Uncertain values are flagged for a person, never silently accepted — that is the behaviour to evaluate.
We publish no accuracy percentages; recognition is assessed on a representative sample of your own printed and handwritten forms during evaluation.
Capabilities
Everything below is confirmed behaviour of the module. Where a capability depends on your documents, the condition is stated.
Text and named fields are read from the scanned page, in Arabic, English, or both mixed on one form. Printed and handwritten content are both supported.
Values the recognition is not confident about are marked, not silently accepted. Handwriting and poor scans raise flags more often — as the demonstration shows.
A reviewer sees each flagged value next to its location on the original page, and confirms or corrects it. Nothing moves on without that decision.
The scanned document is kept with the record, so any value can be traced back to the page it came from — now and during later audits.
Only reviewed, structured information moves on into the FormXS route. Validation, calculations and approval rules apply from there.
The original reading, the reviewer's decision and the final value all land in the record's audit history, with who did what and when.
Who can see a document, review a value or open a record is governed by the same role-based access model as the rest of Data X.
Handwriting recognition accuracy depends on the handwriting, the scan quality and the document type. Difficult handwriting means review does more of the work — by design.
In Data X
OCRXS feeds the records layer that the rest of Data X works on. A reviewed record enters the FormXS route — validation, calculations, approvals and generated documents — and, once approved, becomes something authorised users can ask XVS.ai about.
The three modules share one permission model and one audit history. Each is available on its own site; together they are Data X, by Grid Applications Ltd.
Hover or focus a node to see what it does, or play the record PR-2026-00147 through the platform.
Deployment and governance
Scanned documents are among the most sensitive material an institution holds. OCRXS runs where Data X runs — and that is decided by you.
Data X is deployed on-premises or in an approved sovereign private-cloud option. Hosting, access, backup, recovery and support arrangements are defined per deployment.
Who: the deployment owner. What: the chosen hosting shape and who may administer it. When: at deployment, and at every change.
Data X works with the systems you already run through approved APIs. Complex integrations and specialised requirements may require technical work, and the boundary is made explicit during scoping.
Who: the connecting system’s identity. What: which records it read or wrote through an approved API. When: each exchange.
Documents, extracted values and records remain the institution's. Documents and extracted values are processed where Data X is deployed — on-premises or in the approved sovereign private cloud.
Who: the account that opened a document. What: which scanned page and which extracted values were viewed or exported. When: each access.
Role-based access governs who can see and review documents. Every reading, review decision and final value is written to the audit history.
Who: the reviewer. What: the original reading, the decision and the final value. When: at every reading, review and hand-off.
Regulatory fit is assessed with your own technical and compliance teams against your obligations; no compliance status is claimed here.
Where it is used
Example processAccount-opening mandate
Decisions on the account-opening mandate remain with the institution’s people, in the route they define.
OCRXS proposes structured values and keeps the source. Data X never independently makes lending, clinical, legal or regulatory decisions — those remain with your people, in the route they define.
Questions
Handwritten recognition is supported in both Arabic and English, with accuracy that genuinely varies by handwriting and scan quality. Values the system is unsure of are flagged for human review — for difficult handwriting, expect review to do more of the work.
It is retained with the record. Reviewers see extracted values beside their location on the page, and anyone auditing the record later can open the source document.
No. Recognition only proposes structured values; flagged values wait for a reviewer, and approvals always follow the human route defined in FormXS.
Because a single percentage would not describe your documents. Clean printed forms and rushed handwriting on photocopies behave very differently. Accuracy is evaluated on a representative sample of your own forms, and the results are reviewed together.
With your documents: real, anonymised forms from the actual process, both languages, printed and handwritten variants, good scans and bad ones. The practical questions are which documents it reads well, how often review intervenes, and whether that beats full re-typing for your volumes. Evaluation documents are shared only through the secure channel agreed for the engagement.
Nowhere outside your deployment. Scanned pages and extracted values are processed where Data X is deployed — on-premises or in the approved sovereign private cloud — and stay with the record under the same roles and audit history.
Contact
A demonstration walks through capture, review and the hand-off into the workflow, using your document types where possible. Tell us what you receive on paper today.