MaLLaM LLM: Mesolitica Builds Malaysian Large Language Model for Generative AI Assistants on AWS
Leading Malaysian AI Startup Builds Culturally Relevant Model That Understands More Than 16 Languages and Dialects, Enjoys 5.5-Fold Increase in Throughput While Reducing Compute Costs By 87% at 20% Lower Latency
Amazon Web Services (AWS), an Amazon.com company, has announced that Mesolitica, a Malaysian startup specialising in training large language models (LLMs), has built a Malaysian language generative Artificial Intelligence (generative AI) LLM on the world’s leading cloud. The MaLLaM LLM can understand local nuances like slang, colloquialisms that merge different dialects, Bahasa Malayu, and 16 other regional languages for use in AI assistants across industries.
Mesolitica has trained MaLLaM LLM on 197 datasets totaling close to 200 billion tokens of publicly available Malay-specific content to provide culturally-relevant AI support for applications in customer service, content generation, and data analysis in localised languages.
Using custom ML chips, including AWS Trainium, and AWS Inferentia, Mesolitica saw compute cost savings of 87% while enjoying a 5.5-fold increase in throughput (transactions per second) while training MaLLaM LLM, improving the model’s responsiveness and efficiency when used for AI assistants. With AWS, Mesolitica can now deploy proofs of concept (PoCs) in as little as 24 hours, while the AWS Asia Pacific (Malaysia) Region provides a 20% reduction in latency, critical for achieving human-like conversations in AI voice assistants.
“Our biggest challenge was understanding the many local languages Malaysian patients use,” said Dr Kev Lim, CEO and Founder of health-tech startup Qmed Asia. “By leaning into Mesolitica, we can now better understand local speech patterns through MaLLam LLM. This has enhanced the accuracy of our medical note-taking solution, strengthened patient communication, and ultimately, empowered us to deliver higher quality healthcare.”
Real-World Uses of MaLLaM LLM
With AWS, Malaysian enterprises using MaLLaM LLM can improve operations with generative AI in regional languages to help underserved audiences like farmers in rural areas make data-driven decisions using real-time weather forecasts, soil health analysis, and crop viability assessments.
The Malaysian government is also exploring the integration of MaLLaM LLM into its operations, which aligns with the country’s broader goal of AI sovereignty and local data governance. AI assistants built on MaLLaM LLM can provide quick, accurate responses to citizens’ inquiries in multiple languages, including dialects from different Malaysian States such as Johor, Kedah, Sarawak, Selangor, and Terengganu, to ultimately improve citizen communication and data processing capabilities across the culturally diverse country.
Malaysia’s educational sector can benefit from MaLLaM LLM through applications in language learning and research, particularly in enhancing the understanding of local languages and dialects.
“With AWS, we can deploy proofs-of-concept much faster, with the right cost-effective AI compute resources and machine learning capabilities,” said Khalil Nooh, Co-Founder of and CEO at Mesolitica. “This allows our customers to focus only on the ongoing operational costs, rather than upfront capital expenses, for their AI experiments. This is also in line with Malaysia’s national priority to develop citizen-centric applications, making our MaLLaM LLM generative AI assistant strategically important to the country’s digital transformation ambitions.”
Navigating a Region of Diversity with AWS Capabilities
Southeast Asia’s population speaks about 2,300 languages. When LLMs that are pre-trained in English and western-centric data are tasked with non-English queries, they can produce inaccuracies and misinterpretations. LLMs that are trained on culturally diverse data like MaLLaM LLM can address this gap, boost accuracy, and better cater to the region’s diverse cultures, ways of working, and languages. As a cloud-native platform, Mesolitica needed compute-intensive resources to develop local LLMs and meet demand from private and public sector customers across the country. 
Mesolitica has significantly enhanced its Machine Learning operations by leveraging AWS services. The company migrated its model training workloads to Amazon Elastic Cloud Compute (Amazon EC2) and deployed inference workloads using Amazon EC2 G5 instances, which provide cost-effective GPU acceleration for AI models. To enhance its infrastructure, Mesolitica implemented Amazon Elastic Kubernetes Service (Amazon EKS) to deploy and manage ML models and applications, and to orchestrate P4 Nvidia instances. Furthermore, the company utilizes Amazon SageMaker, a fully managed ML service, to efficiently manage and prepare large data sets essential for training LLMs.
Mesolitica is part of the AWS APJ Generative AI Spotlight program, a four-week accelerator program which aims to support early-stage startups in the region that are developing generative AI applications. The startup is also one of two Malaysian companies to receive AWS credits from the AWS Activate Program, a comprehensive initiative by AWS that provides access to a range of resources, including AWS credits, technical support, training, and tools tailored to help startups build, launch, and scale their applications on AWS. Mesolitica has joined the AWS Partner Network, a global program designed by AWS to assist businesses in leveraging AWS for growth and success.
“Malaysian startups can revolutionise industries through data-driven solutions and advanced technologies,” said Pete Murray, Country Manager at AWS Malaysia. “For GenAI to be relevant, it must be accessible and culturally integrated. Mesolitica is creating Malaysia’s first Large Language Model AI that’s tailored to the country’s diverse population. This has potential to support various sectors, from improving government services to financial inclusion. We’re proud to host MaLLaM LLM on AWS Malaysia region.”
