Thammasat students develop AI for ‘ThaiKALA’, a Thai language question answering tool

A conversation with Mr. Pawarit Ruangjutiphophan, a fourth-year student majoring in Computer Engineering, Information and Communication Technology, at the Sirindhorn International Institute of Technology (SIIT), Thammasat University after receiving the first place, a gold medal, in the “Super AI Engineer Season 3” competition, a project on developing talents in the fields of inventor-engineer-researcher-entrepreneur. His work has also been published as a paper recently.

Details of the project

It is the AI Training Camp project which aims to collaborate with various organizations to solve challenges through a Hackathon competition. The camp runs for a duration of one year, with weekly Hackathon events. Participants are divided into teams (in selecting houses) and can choose the house they want to join. The chosen house is the team that participants work together to solve Hackathon challenges from the beginning until the end of the project.

Thai version of ‘ThaiKALA’ ChatGPT

The work is called ‘Thai Knowledge Augmented Language Model Adaptation’ (ThaiKALA). The idea originated from the fact that AI is being widely used nowadays. One AI that we frequently encounter is a technology called Language Model, with an example being ‘ChatGPT’, which answers various questions in a way that resembles human responses. However, these models are commonly developed with English as their primary language, making it difficult when it comes to answering questions in Thai. Consequently, a large amount of data is required to construct a model that can effectively operate with the Thai language.

My work utilizes a Knowledge Graph, which stores various interconnected datasets in the form of a Graph, to enhance a small Language Model currently in use. This is achieved through a framework called ‘ThaiKALA’, which we have developed. The purpose is to enable the model to access knowledge from the Knowledge Graph and utilize it to assist in answering various questions.

This work demonstrates the exceptional capabilities of using Knowledge Graph data in a small-sized Language Model. It shows that even though the model is smaller in size compared to ChatGPT, it can still generate responses to Thai language questions with comparable abilities. This work was conducted during my internship at Kasikorn Business Technology Group (KBTG) and has been published as a paper. For more information, please read the article at https://ieeexplore.ieee.org/document/10355001.

 Fundamental knowledge

For those who are interested in competing in the field of technology or Hackathon, if you want to prepare yourselves as competitors, having a good fundamental knowledge will greatly enhance the abilities. The curriculum at SIIT covers important fundamentals, and personally, I never missed a class and paid attention to the professors’ teachings. I always asked questions whenever I had doubts and actively participated in group work with my peers, giving my best effort every time. These things train the fundamentals of working with others. Although these may seem trivial, these small fundamentals are extremely important in competition.