Invoices and receipts
Read purchase invoices, receipts, and supplier statements and move fields into the finance posting queue.
AI services / documents
Document automation and Persian OCR with AI
Turn Persian invoices, contracts, forms, and correspondence into structured data and deliver every item to your existing workflow with a human approval point.
OCR versus intelligent document processing
OCR only converts an image to text. Intelligent document processing goes further: it classifies the document, extracts meaningful fields (amount, Jalali date, national ID, contract number) with confidence scores, validates them against reference data, and routes ambiguous cases to a specialist. This service designs both layers.
Documents that are usually worth automating
Contracts and amendments
Extract parties, amounts, dates, obligation and penalty clauses for a searchable archive.
Administrative forms
Registration, request, insurance, and case forms with fixed fields, signatures, or stamps.
Correspondence
Classify incoming letters, assign the responsible unit, and draft a reply for review.
Identity documents
Extract card, certificate, and credential data with privacy controls and limited retention.
Scanned archives
Digitise paper archives and build a semantic search layer on top.
How we work
Extraction accuracy is measured on a real sample from your organisation, not on a prepared demo. Scan quality, handwriting, stamps, and layout variation have the largest effect on results.
- 01
Sampling
A representative sample of each document type, including poor and exceptional cases, is prepared with unnecessary data removed.
- 02
Layered extraction
A combination of Persian OCR, a language model, and validation rules (for example line totals matching the grand total) per field.
- 03
Confidence scoring
Each field carries a confidence score; below the threshold it goes to the exception queue, not to automatic posting.
- 04
Workflow integration
Structured output is delivered to a controlled file, work queue, or the software interface of the current system.
- 05
Measurement
Per-field accuracy, specialist correction rate, and processing time are compared with the manual baseline.
Design controls
- Exception queue
- Low confidence, incomplete documents, and conflicting data are routed to a specialist rather than guessed.
- Decision trail
- The original image, extracted field, user edit, and final approval are recorded for audit.
- Privacy
- Personal data is removed or masked and retention follows the organisation's policy.
- Periodic evaluation
- As document templates change, accuracy drift is tracked with approved samples.
Zharf
Live demoTry the ZharfAI assistant right now
Ask about Persian assistants, document automation, or how to start a pilot.
Frequently asked questions about document automation
How accurate is Persian OCR?
There is no fixed number; accuracy depends on scan quality, font or handwriting, stamps, and layout. That is why we report per-field accuracy on your real sample, and critical fields are always protected by a confidence threshold and human review.
Is Persian handwriting supported?
Handwriting can be evaluated but its error rate is higher than printed text. For handwritten fields we usually recommend a combination of automatic extraction and human confirmation, and the pilot shows which fields are worth automating.
Which systems can receive the output?
Structured output (JSON or CSV) can be delivered to an accounting system, office automation, an archive, or a work queue. The connection type is decided after reviewing the current system and access requirements.
Does document data leave the organisation?
The deployment model is assessed against each organisation's security requirements; processing on your own infrastructure or hosting inside Iran is available, and the data flow is documented before the project starts.
How long does it take?
A bounded pilot on one document type usually takes a few weeks and ends with a report on accuracy, exception rate, and processing time. Expansion to other documents follows a decision based on that report.
Choose one document type for a pilot
Specify the document type, monthly volume, current handling time, exceptions, and approval point. Discovery may lead to a pilot or a clear decision not to proceed.