Riyadh-based Misraj AI just solved a problem that affects millions of Arabic speakers across the Middle East. The company launched Kawn, an AI system that understands more than a dozen Arabic dialects with native fluency, at AWS re:Invent 2025 event.
Most AI systems struggle with Arabic because they’re built for English first. Companies across Saudi Arabia, Egypt, and the UAE have been stuck using AI tools that can’t properly read their business documents or understand how their customers actually speak. Misraj AI changed this by creating AI from the ground up for Arabic speakers.
Kawn Ecosystem Breaks Language Barriers Across the Arab World
The Kawn ecosystem includes several AI tools that work together. The main language model handles formal Arabic and regional dialects with equal skill. Its Lahjawi engine recognizes Gulf, Levantine, Egyptian, and North African dialects without confusion.
Safwan AlModhayan, CEO of Misraj AI, explains the core problem they’re addressing: “Most existing AI models are originally built for English and later adapted to Arabic, which results in limited accuracy, weak dialect support and unreliable performance in important sectors such as government, healthcare, finance and education”.
The company developed a technique called layer injection that lets the model learn new dialects without retraining the entire system. This approach means businesses can add support for their specific regional language without waiting months for updates.
Three Major Tools Launch Together
Misraj AI announced three interconnected products at the Amazon Web Services conference:
Kawn Ecosystem provides the foundation with large language models, vision-language models, and retrieval systems designed specifically for Arabic. Unlike adapted English models, Kawn was trained on over 2 trillion Arabic tokens from diverse regions and industries.
SeamlessAPI gives developers a single interface to access all Arabic AI capabilities. One API call can translate text, detect dialects, summarize documents, or extract structured data from PDFs. This simplifies integration for companies that don’t want to manage multiple AI services.
Baseer converts scanned Arabic documents into clean, AI-ready text. The tool preserves document layout while supporting multilingual content. This matters for organizations with large paper archives that they want to digitize.
Saudi Vision 2030 Drives Arabic AI Investment
Saudi Arabia’s Vision 2030 initiative includes major investments in AI technology that serves Arabic speakers. The kingdom launched HUMAIN, another Arabic language model, and created the Saudi Data and AI Authority to lead regional AI development.
This focus on Arabic-first AI development addresses a significant market gap. Arabic is spoken by over 400 million people globally, but most AI systems treat it as a secondary language. Companies across the Middle East have struggled to extract value from Arabic documents and serve Arabic-speaking customers effectively.
The economic impact extends beyond Saudi borders. Gulf states, North African countries, and Levantine nations all face similar challenges with AI tools that can’t handle their specific dialects or business contexts.
Enterprise Applications Drive Adoption
Early use cases show Kawn’s practical value for businesses. Banks can now analyze Arabic loan applications automatically. Government agencies can process citizen requests in local dialects. Healthcare systems can understand patient records written in different Arabic varieties.
The Workforces platform, also launched at AWS re:Invent, lets companies create AI agents for specific Arabic workflows. These agents can handle customer service in multiple dialects, analyze Arabic market research, or automate compliance processes for regional regulations.
Insurance companies represent a key target market. They hold millions of Arabic documents that current AI can’t process reliably. Kawn’s document intelligence capabilities let them digitize claims, policies, and reports while maintaining accuracy across different Arabic writing styles.
Technical Innovation Addresses Core Challenges
Building effective Arabic AI required solving several technical problems. Arabic text flows right-to-left, uses different scripts for the same letters depending on position, and includes diacritical marks that change meaning.
Dialectical variation posed the biggest challenge. A customer service chatbot needs to understand whether someone from Cairo says “عايز” (ayez) or someone from Riyadh says “أبغى” (abgha) – both meaning “I want” but in completely different dialects.
Misraj’s layer injection technique lets the model handle these variations efficiently. Instead of training separate models for each dialect, Kawn learns dialectical patterns as additional layers on top of its core Arabic understanding.
Regional Expansion Plans Target Growing Demand
AlModhayan outlined ambitious expansion plans beyond the initial launch. The company will develop sector-specific models for healthcare, legal services, insurance, and education. Each vertical requires specialized vocabulary and compliance understanding.
Multimodal capabilities represent the next development phase. Combining text, speech, and vision lets Kawn handle video calls in Arabic dialects, analyze Arabic documents with charts and images, or provide voice responses in regional accents.
The company aims to become the primary Arabic AI foundation across the Middle East and North Africa. This requires partnerships with local integrators, government approvals, and customization for each market’s specific needs.
International expansion could follow regional success. Arabic-speaking communities in Europe, North America, and Australia need AI tools that understand their language and cultural context.
