Pathumma - Thailand’s AI Model aiming for Agentic AI with strategic partnerships

Amid the rising popularity of Large Language Model (LLM) technology—advanced AI models capable of human-like language processing and communication—the National Electronics and Computer Technology Center (NECTEC) under the National Science and Technology Development Agency (NSTDA) has developed Pathumma LLM, a Thai Large Language Model, to support the advancement of AI services in the country.

Sarawut Kongyang, a researcher in the Artificial Intelligence Research Group at NECTEC-NSTDA, explained that LLMs can be categorized into two types: closed models, where developers do not allow public access for further development—such as ChatGPT, Gemini, and Claude—and open models, which are publicly available for further development—such as Gemma, Sea Lion, Typhoon, and THaLLE.

Pathumma LLM has been developed as an open model to address the limitations of closed models, which may lack flexibility, particularly for organizations that require data confidentiality, such as banks, healthcare institutions, and courts. It also caters to Thailand-specific applications, such as drafting official government documents and communicating in regional Thai dialects.

"The research team leveraged its expertise in AI model development alongside NSTDA's internationally recognized infrastructure, including the LANTA supercomputer, to build Pathumma LLM. The first version, 1.0.0, is now available for trial on the AI For Thai website. Interested users can access it in three formats:

APP – A web application for general users.

API – For developers who want to integrate it into their websites.

Model – For developers who wish to download and further develop the model."

A key feature of Pathumma LLM is its Multi-Modal Generative AI capability, which supports three types of data processing:

Text LLM – A model designed for processing text-based queries and commands, fine-tuned for information retrieval and logical reasoning. This makes it ideal for enterprise-specific applications. For example, the Revenue Department could use it to develop a chatbot that answers tax-related questions for the public, while research institutions could use it to create a chatbot for searching and summarizing organizational research data.

Audio LLM – A model optimized for processing audio data. It can transcribe speech in both Thai and English, generate environmental sound descriptions, detect the speaker’s emotion and gender, and answer questions or provide insights based on the content of an audio clip.

Vision LLM – A model designed for processing image-based data. It is fine-tuned to generate image descriptions, extract text from images, and answer questions or provide insights related to the images.

However, enhancing the efficiency and capabilities of LLMs requires multiple factors, particularly data volume, quality, and relevance, as well as a robust infrastructure, such as cloud computing systems for AI processing. To address this, the research team plans to collaborate with both public and private sector partners in Thailand to develop a larger foundation model tailored for the country. This initiative aims to expand the model’s capacity to handle more data and parameters for AI training, with implementation expected to begin by February 2025.

The research team also plans to request publicly available data from Thai government agencies and private organizations to train the AI. This will help Pathumma LLM build a robust knowledge base, making it an open model with expertise in Thai language, information, and context. Additionally, the model is expected to contribute to AI Governance, enhancing the efficiency of government services through AI-driven solutions in the future.

"Our next development goal for Pathumma LLM is not only to strengthen its data capabilities and Generative AI functions but also to advance it toward becoming Agentic AI—an AI capable of autonomous reasoning and decision-making. For instance, it could serve as a personal AI assistant that analyzes market trends and executes stock trades based on predefined conditions, or as a smart chatbot that recommends services tailored to user preferences while automatically tracking task statuses. The shift toward Agentic AI is a key global technology trend, and our team is committed to driving Pathumma LLM in this direction."