On this morning, the URA Research Group opened its doors to a delegation from VNG Corp. in the conference room at Ho Chi Minh City University of Technology. Leading URA’s side was Assoc. Prof. Quản Thành Thơ, who first founded the lab in 2009 and rebranded it URA in 2021, alongside several senior researchers. VNG arrived eager to understand URA’s research activities, assess collaboration potential, and identify the difficult problems during researching and working.
URA kicked off the session with a concise history: since June 2009 the group has grown to more than 100 active members (five faculty, five research fellows, and over ninety undergraduates). Its core focus areas include:
- Large‑scale Vietnamese language models
- Computer vision (applications in healthcare diagnostics, industrial inspection, and education)
- Time‑series analysis (from battery health monitoring to epidemiological forecasting)
- Intelligent reasoning systems (advanced Q&A agents with inference capabilities)
The senior URA researchers then highlighted a range of flagship projects—from an end‑to‑end AI agent for Vexere.com that seamlessly handles booking inquiries, ticket changes, and payments across Facebook and Zalo to the HCMUT Chatbot powered by URA’s own Vietnamese LLM, which personalizes recruitment advice about courses, policies, and campus life. They also demonstrated a Virtual Try‑On System that maps garment contours to user poses in real time for an immersive online shopping experience, an MRI‑Based Knee Degeneration Detector that uses deep learning to highlight damaged regions, quantify severity, and display confidence scores for clinicians, and a geographic time‑series model for COVID‑19 forecasting that projected infection and recovery scenarios to inform local health authorities.
URA faces several critical challenges when translating research into industry‑ready solutions. First, the scarcity of high‑VRAM GPUs (≥24 GB) forces us to downscale models or reduce experiment counts, prevents benchmarking promising research directions, and disrupts student training when they must queue for shared hardware—directly undermining our goals for international publications and timely product delivery. Second, slow funding disbursement—even for approved grants from the university, research funds, or corporate sponsors—means we often miss the window to rent cloud servers or onboard student researchers at crucial moments, eroding the flexibility that Agile sprints demand. Finally, lengthy internal approval processes for additional resources (whether extra storage, RAM upgrades, or new GPU access) leave our research plans “on ice,” delaying external collaborations and hindering the rapid infrastructure decisions essential for sustained high performance.
We extend our sincere thanks to the VNG team for their time, insightful questions, and willingness to explore partnership opportunities. We look forward to transforming today’s discussions into concrete collaborations and are excited to embark on joint projects that will advance AI research and its real‑world impact in the near future.