Recently, the Saudi Arabian AI laboratory Misraj AI launched its new Arabic large language model Kawn at the AWS re:Invent conference. The model aims to promote Arabic communication across industries, offering a variety of models and services to support Arabic applications.
An important feature of the Kawn model is Mutarjim, a bidirectional Arabic-English translation model. In addition, Lahjawi is the first translation model that supports 15 Arabic dialects, making cross-dialect communication more convenient, especially suitable for use in chatbots and customer support scenarios. Kawn also includes Kawn Document OCR, a vision-language model that can recognize and extract Arabic text from documents.
Image credit note: The image is AI-generated, and the image licensing service provider is Midjourney
Misraj AI's CEO Safwan AlModhayan stated that the launch of Kawn aims to fill the gap in high-quality local Arabic AI systems in the market. He pointed out that many existing AI models were originally built for English and then adapted for Arabic, leading to shortcomings in accuracy, dialect support, and reliability in important fields such as government, healthcare, finance, and education.
To build Kawn, the team faced many challenges, including how to design a model that could cover the wide range of Arabic dialects. To address this, Misraj developed a technology called layer injection, which allows the model to learn dialect variations without retraining the entire model. Another challenge was the lack of clean and representative Arabic datasets. To tackle this issue, the team collected and refined over 2 trillion Arabic language tokens from multiple fields and regions, and established strict filtering and quality control processes.
In addition to Kawn, Misraj also launched Workforces, an AI platform that allows users to create and manage AI agents to automate workflows, analyze data, and expand business. The platform is suitable for operations, customer support, marketing, and sales teams, aiming to help businesses automate repetitive processes and improve productivity.
AlModhayan said that Kawn and Workforces are intended to usher in a new era of Arab digital transformation. They hope to help institutions and companies in various industries deploy AI solutions that can naturally use Arabic, understand regional processes, and integrate seamlessly with existing systems. Looking ahead, Misraj plans to further expand the platform by launching specialized models for industries such as healthcare, law, insurance, business, education, and public services. At the same time, they are also developing richer multimodal capabilities, combining text, voice, and vision, to provide users with a more interactive experience.
Key points:
🌟 Kawn is the first large language model in Arabic, supporting Arabic communication and translation.
🔍 The model overcomes the challenges of Arabic dialect diversity and the scarcity of datasets.
🚀 Misraj also launched the AI platform Workforces, helping enterprises automate workflows.
