Arabic OCR
Extract text from PDFs and scanned images while accounting for page layout and source quality.
Arabic-first document intelligence
Restore Arabic documents.
Turn them into verifiable knowledge.
Madarek reads scanned PDFs and images, repairs common Arabic extraction defects, and builds a searchable knowledge layer that answers from source pages with clear citations.
Prototype in active development · Riyadh, Saudi Arabia
استُخرج هذا النص من ملف ممسوح ضوئيًا
فتعود الكلمات إلى نسقها، والحروف إلى وصلها
Arabic-first pipeline
Page-level citations
Semantic and hybrid search
Open integrations
Flexible deployment
Capabilities
An end-to-end Arabic document pipeline, from scanned page to verifiable answer.
Text restoration
Repair disconnected, reversed, and disordered Arabic text, then recover headings, paragraphs, and tables.
Semantic retrieval
Combine semantic, lexical, and reranking methods to retrieve relevant evidence more accurately.
Citation-backed answers
Link generated answers to the exact document, page, and evidence segment used.
Privacy by design
Support customer-hosted or isolated private-cloud deployment based on data sensitivity and policy.
Automation and APIs
Process document batches and connect structured outputs to knowledge tools, agents, and products.
How it works
A controlled path from raw file to searchable, reusable knowledge.
- 1
Ingest
Upload a file or submit it through an API into an isolated workspace.
- 2
Read
Select the appropriate extraction path for each page type and quality level.
- 3
Restore
Normalize direction, repair text, and recover document structure.
- 4
Index
Segment content, generate representations, and build a retrieval index.
- 5
Use
Search, ask, export, or expose the knowledge to assistants and agents.
Deployment that matches data sensitivity
Madarek is designed to support multiple operating models for security, sovereignty, and scale.
Arabic documents deserve to be understood.
Madarek turns unstructured Arabic content into trustworthy, searchable, and interoperable knowledge.
Request early accessPrototype · Open to pilots and technical partnerships
Contact
For early access, technical partnerships, or an Arabic document pilot.