Built, hosted and supported inside the Kingdom. From the infrastructure in the data centre to the model that reads the page.
Contracts, title deeds, commercial registrations, permits, court filings, engineering drawings, material test certificates, attestations and personnel files. Decades of them. Organisations here are digitising at remarkable speed, but what they are producing is images of paper. A scanned deed is still unreadable to any system that needs to act on it.
Handwriting, official stamps and seals, Hijri dates, right to left tables and mixed language pages. The global document platforms were trained on English forms and fail on all of it.
Data residency rules and the Personal Data Protection Law prevent government, banking, legal, energy and healthcare entities from using foreign clouds at all.
Digitisation produces volume, not usability. The gap between how much is stored and how much can actually be queried widens every year.
Forged stamps, altered dates and fake certificates move through procurement, lending and licensing. Detection is still a person reading pages one at a time.
Most companies in this category rent someone else's cloud and fine tune someone else's model. Neither option is open to a customer who cannot let a document cross a border.
Tasjeel AI operates its own data processing and hosting infrastructure inside the Kingdom, builds the Arabic document intelligence that runs on it, and maintains the applied research that keeps both ahead. From the metal in the rack to the model that reads the page.
The company exists to turn the Kingdom's registration culture into trustworthy data. Every document in an archive is a record of something that was registered. We make that record legible to a machine, and provable to an auditor.
Each layer stands on its own. A customer can take the cloud alone, the reader alone, or the whole stack as a single managed deployment inside their own facility.
A data science workbench over the full record graph. Query across every document an organisation holds, trace a clause or a counterparty through decades of filings, and build reports on evidence rather than recollection.
Validation and compliance. Checks each record against the rules that govern it, flags missing or expired documents before an auditor does, and produces the evidence trail that proves a file was complete on a given date.
The structured record store. Versioned, with full lineage from extracted field back to the pixel it came from, a complete audit trail, and an API that an organisation's existing systems register against.
The Arabic document intelligence engine. Handwriting, official stamps and seals, Hijri dates, right to left tables, degraded and skewed scans, and forensic detection of tampering and forgery.
In-Kingdom hosting and data processing infrastructure. Everything above runs on it, and it is sold directly as sovereign cloud and managed hosting to organisations that need capacity without the compliance question.
Five stages, each leaving evidence behind. Nothing in the pipeline requires the document to leave the customer's jurisdiction, and every extracted value stays linked to the exact region of the page it came from.
Scans, photographs, PDFs and native files arrive from an archive, a scanner or an existing system. Pages are deskewed, denoised and classified by document type.
Qari reads the page. Printed and handwritten Arabic, English, mixed layouts, stamps, seals, signatures and tables, each located and transcribed with a confidence score.
Seals and signatures are checked against known patterns. Tampering, altered dates and inconsistent impressions are flagged for human review rather than passed silently.
Values become a structured record in Sijil, versioned and linked back to the source region on the page, so every field can be defended by showing where it came from.
Mizan validates the record against the rules that apply to it, Diwan makes it queryable, and the API hands it to whatever system needed it in the first place.
Arabic document understanding is an open research problem, not a solved one waiting to be bought. Labs is co funded with universities and government bodies, and every finding feeds directly back into the products rather than into a paper nobody implements.
Printed, handwritten and historical forms. Ligature heavy scripts, inconsistent baselines, marginalia, and the degraded scans that make up most of what is actually in an archive.
Seal and stamp verification, tamper and splice detection, ink and impression analysis, and the provenance signals that distinguish a genuine certificate from a convincing one.
Machine reading of piping and instrumentation diagrams, isometrics, as built drawings and material test certificates, where the information lives in geometry as much as in text.
Each one holds large archives, strict residency requirements and a direct financial cost attached to records that cannot be trusted.
Contracts, filings and case files. Decades of precedent locked inside scanned bundles that cannot be searched, cited or cross referenced.
Deeds, lease registries and owners association records, where a single misread boundary or date carries direct financial consequence.
Drawings, as builts, quality records and material certificates. The most technically demanding vertical, and the one where almost nobody else is competing.
Know your customer files, commercial registration verification and trade finance documents, where verification speed is the constraint on the business.
A body of Gulf documents and Arabic handwriting that no global vendor holds, and that residency law prevents them from legally acquiring.
Infrastructure inside the Kingdom plus the cloud service registration that goes with it. A barrier a competitor has to cross, not buy.
Every document processed improves extraction and deepens switching cost. The platform is worth more to a customer in year three than in year one.
The hardest capability in this category to replicate and the easiest to sell, because its value is measured in fraud that did not happen.
If you hold an archive that cannot leave the country and cannot stay as paper, we would like to hear about it.