Invoice OCR
Supplier, tax number, dates, line items and totals — with the arithmetic checked, so a total that does not match its lines is flagged rather than delivered.
- Line items
- Tax number
- Totals reconciled
- Multi-currency
SLT OCR · a Smart Lead Tech product
Turn your organisation's documents into data you can trust
An Arabic-first computer-vision engine for document process automation: invoices, purchase orders, receipts, passports, identity cards, KYC packs, and mixed multi-page PDFs. Every field has its own verification model, and values that cannot be verified are routed to human review rather than delivered as fact.
- A dedicated model for every field
- Human review when needed
- Processed in memory only
- Cloud or on-premises
97.9%page classification accuracy on sorted held-out documents
3validation layers before any value is delivered
15 minuntil every uploaded result is erased
0documents written to disk at any stage
Built for Arabic
An Arabic-first document understanding solution
Arabic is not a language you bolt on at the end. Right-to-left layout, letters that change shape with position, Arabic-Indic numerals, and handwriting that varies by governorate all break engines designed around Latin script. SLT OCR was trained on Arabic documents from the start, on real archives rather than synthetic text.
- Right-to-left layout and mixed-script pages handled natively
- Arabic-Indic and Eastern Arabic numerals normalised automatically
- Trained on real scanned archives, not generated samples
- Every extracted value carries its own confidence and evidence
How it works
Three stages, and a refusal to guess
Read the page
Each page is classified by type before anything is extracted, so an identity card is never processed as an invoice. Pages that do not fit a known type are flagged rather than forced.
Extract each field
Fields are resolved by dedicated models in dependency order — the identity number is parsed and checksummed before the values derived from it are trusted.
Verify, or escalate
Structural checks, cross-field agreement and confidence thresholds decide whether a value ships. Anything below the bar goes to a human queue with its evidence attached.
Ready-made solutions
Break free from manual data entry with ready-made document understanding
Each pipeline is trained for one document family and ships with its own validation rules. Start with the one that matches your backlog.
National ID OCR
Front and rear of the same card are recognised as one person and merged into a single complete record. The number is checksum-validated and the derived fields cross-checked against it.
- Front + rear merged
- Checksum validated
- Derived fields
- One row per person
Passport OCR
The machine-readable zone is read and its check digits verified arithmetically, then reconciled against the printed page. A mismatch is a review item, not a silent choice.
- Machine-readable zone
- Check digits
- Document dates
- Nationality
Purchase Order OCR
Order numbers, quantities and delivery terms lifted from supplier paperwork and matched against the order they belong to.
- PO matching
- Quantities
- Delivery terms
- Approval routing
Receipt OCR
Photographed, folded and faded receipts — merchant, date, tax and total, from the awkward originals people actually submit.
- Camera photos
- Merchant
- Tax split
- Expense codes
KYC Automation
A folder of mixed identity evidence resolved into one verified customer profile, with every field traced back to the page it came from.
- Mixed evidence
- Identity resolution
- Audit trail
- One profile
PDF Page Classification & Bookmarking
A scanned PDF with no structure goes in; the same PDF comes out with a bookmark tree naming each document inside it. Pages need not be in order.
- Per-page typing
- Bookmarks written
- Shuffled input
- Content unchanged
Intelligent Table Recognition
Rows, columns and merged cells recovered from scanned tables, including the ruled Arabic forms that defeat generic layout models.
- Merged cells
- Ruled forms
- Column alignment
- Export to Excel
Image-to-Text OCR
Skewed camera captures, low-contrast photocopies and stamped pages — deskewed, cleaned and read.
- Deskew
- Low contrast
- Stamps and overlays
- Mixed script
Where it is used
Industries that run on paperwork
Government
Citizen records, archives and licence files digitised without leaving the premises.
Energy and utilities
Contract archives, meter records and customer files at box-by-box scale.
Banking and finance
KYC packs, statements and onboarding evidence with a traceable audit trail.
Healthcare
Patient forms and insurance paperwork, processed without documents leaving the network.
Legal
Deeds, contracts and case bundles made searchable and correctly bookmarked.
Logistics
Delivery notes, customs paperwork and proof-of-delivery captured at the depot.
Why SLT OCR
Engineering decisions you can inspect
Runs where your data lives
Cloud, or fully on-premises with no outbound connection. The on-premises build works air-gapped.
Nothing written to disk
Uploads are processed in memory and results expire automatically. There is no document store to breach.
Confidence, not assertions
Every field carries a confidence and its evidence. Low-confidence values are labelled, never quietly guessed.
Arabic as a first language
Trained on real Arabic archives, not translated Latin corpora.
One row per person
Fragments of the same identity across several files are merged, so counts are correct.
Measured, and published
Accuracy figures state the input they were measured on. Where a scenario is weaker, we say so.
Try it
Run a document through the engine
Upload a document and see the extracted fields, their confidence, and what would have gone to review. Trial processing is rate-limited and results are erased automatically.
Contact
Ready to evaluate SLT OCR inside your organisation?
Tell us your document volumes and whether you need an on-premises installation. We will reply with packaging and a deployment plan.