FPT achieves outstanding results in AI applications for Vietnamese language processing at VLSP 2025
On October 29-30, FPT experts took part in the International Workshop on Vietnamese Language and Speech Processing (VLSP 2025), achieving outstanding recognition with two award-winning AI research projects.
This is the most prestigious annual academic event in the field of natural language processing and Vietnamese artificial intelligence, jointly organized by the Vietnamese Language and Speech Processing Association (VLSP), the Institute of Information Technology under the Vietnam Academy of Science and Technology, and leading universities across the country. The workshop gathers a wide range of research institutes, universities, and technology enterprises from Vietnam and abroad, with the goal of advancing AI model development for the Vietnamese community and building a comprehensive Vietnamese language knowledge ecosystem.
FPT wins two top awards at competitions held within the framework of VLSP 2025
Specifically, FPT IS, a member company of FPT Corporation, earned two top awards in competitions held within the framework of VLSP 2025. FPT’s team of experts won First Prize in the LegalQA Challenge – a competition on traffic regulation question-answering based on road signs – and Second Prize in the Deep Retrieval in Legal Documents (DRiLL) Challenge. These achievements not only demonstrate the deep expertise of FPT’s specialists in Vietnamese natural language processing but also reaffirm the company’s pioneering position as a Vietnamese enterprise making its mark on the global technology stage.
Both award-winning studies were presented in poster format at the workshop’s academic exhibition area. These scientific works, which also served as the foundation for the competition results, embody FPT’s application-oriented research approach, closely tied to real-world challenges in Vietnam’s digital transformation journey.
The first study, titled “Data Augmentation and Hierarchical Chunking for Deep Retrieval in the Expansive Legal Landscape”, focuses on addressing the challenges of searching, retrieving, and interpreting legal information within vast and complexly structured data repositories. FPT’s research team proposed a novel approach that combines data augmentation techniques with semantic-based hierarchical chunking, enabling the system to better understand the relationships between legal provisions, sections, and related documents. The results showed that FPT’s model achieved significantly higher retrieval performance compared to traditional methods while notably reducing search time. This solution holds tremendous potential for real-world applications, particularly in the development of digital legal assistants, administrative document search engines, and decision-support platforms for the judicial and public administration sectors – fields that require a high level of semantic understanding and complex document linkage.
The second study, titled “A Low-Cost Low-Energy Approach to VQA on Traffic Signs Problems”, explores a Visual Question Answering (VQA) model – a system that integrates natural language processing and computer vision to enable machines to understand and respond to questions related to traffic sign images. The model developed by FPT stands out for its energy-efficient and cost-optimized design, featuring an architecture that supports centralized processing on servers while allowing edge device access via APIs. This approach delivers significant practical value amid the rapid growth of smart city initiatives, and also opens opportunities to expand applications to other domains such as education, logistics, and traffic safety, where effective image and language understanding under resource constraints is essential.
According to Mr. Pham Quang Nhat Minh, Director of the AI Research and Development Center at FPT IS, FPT Corporation, participation in VLSP 2025 provided FPT’s AI experts with an opportunity to gain hands-on experience, stay abreast of emerging technological trends, and connect with the academic research community in Vietnam. More importantly, it was a chance for FPT to reaffirm its long-term commitment to investing in core technologies, particularly in Vietnamese natural language and speech processing models that serve daily life, governance, and socio-economic development.
“We firmly believe that AI truly delivers value only when it is designed and trained based on local data and context. The research presented at VLSP 2025 marks another milestone in FPT’s journey toward developing ‘Made-in-Vietnam’ AI products that best serve the Vietnamese people,” Mr. Minh emphasized.
As the global AI landscape enters the era of Large Language Models (LLMs), the development of models capable of deeply understanding the Vietnamese language has become increasingly critical. With a consistent strategy to expand its ‘Made by FPT’ ecosystem, FPT Corporation continues to invest heavily in Vietnamese AI research, build a large-scale team of engineers and data scientists, and expand international collaborations to elevate Vietnamese technology on the global stage.
Organized by the Vietnam Informatics Association in collaboration with leading research institutes and universities, and under the auspices of the Ministry of Science and Technology, VLSP 2025 focuses on topics such as large language models, semantic search, speech synthesis, automated question answering, and AI applications in law, transportation, and public administration. FPT’s strong presence at the event not only underscores the research excellence of Vietnamese enterprises but also highlights the company’s role in advancing the national digital transformation and fostering a future where AI truly serves humanity and serves Vietnam.
