Saudi Arabia-based AI lab, Misraj AI, unveiled a “breakthrough” Arabic large language model (LLM) at AWS Re:Invent this week.
llmCON, called Kaavan, offers several models and services to support Arabic communications across industries.
Features of the model include Mutarjim, a bidirectional Arabic-English translation model, and Lahjawi, the first translation model that works on 15 Arabic dialects. The latter enables cross-dialect communication for use cases like chatbots and customer support.
It also includes Kawn Document OCR, which is a visual-language model used to recognize and extract Arabic text from documents.
Safwan Almodhayan, CEO MisrajSaid that the goal is to give businesses and public institutions a “trusted and enterprise-grade” Arabic AI foundation.
“Kavan was created to address a major gap in the market: the absence of high-quality Arabic-native AI systems that can understand the full linguistic and cultural complexity of the Arabic-speaking world,” he told AI Business.
“Most existing AI models are originally built for English and later adapted to Arabic, resulting in limited accuracy, weak dialect support, and unreliable performance in critical sectors such as government, healthcare, finance, and education,” he said.
To create Kaavan, the team had to overcome many challenges, including designing it for the vast range of Arabic dialects spanning regions such as the Gulf, the Levant, and North Africa.
To address this, Misraj developed a technique called layer injection that allows models to learn dialectal variations without retraining the entire model.
“Another challenge was the lack of clean and representative Arabic datasets,” Almodhayan said. “To solve this, we curated and refined over 2 trillion Arabic tokens from a wide range of domains and sectors, supported by rigorous filtering and quality pipelines.”
Along with the con, Misraj also unveiled Workforce, an AI platform that lets users create and manage AI agents to automate workflows, analyze data and scale operations. This will help businesses automate repetitive processes and increase productivity, the company said.
The platform has specialized use cases for teams working on operations, customer support, marketing, and sales.
“Con and Workforce is designed to enable a new era of Arab-first digital transformation“Our goal is to help ministries, regulators, banks, insurers, universities and enterprises deploy AI solutions that work naturally in Arabic, understand regional processes and integrate easily with existing systems,” Almodhayan said.
Looking ahead, Almodhayan said the platform will continue to expand with “sector-specific models” tailored for industries such as health care, legal, insurance, business, education and public services.
“We are also working on rich multimodal capabilities that combine text, speech, and vision for more interactive experiences,” Almodhayan said. “Our long-term vision is for Kaavan to become the primary Arabic AI foundation across the region, giving organizations the tools they need to innovate faster without having to rebuild core AI systems from the ground up.”
