Excellent at algorithms and quick to adapt, yet Vietnamese AI engineers have yet to make a global breakthrough. Join FSI’s CEO for a candid discussion on the barriers, gaps, and opportunities for Vietnam’s engineering force to rise amidst the AI wave.
How would you assess the current capabilities of Vietnamese AI engineers? What are their strengths and weaknesses compared to the regional and global average?
Vietnamese engineers demonstrate strong logical thinking, a solid foundation in mathematics, and a quick ability to self-learn. According to the 2024 TopDev report, Vietnam currently has around 5,000 AI engineers, but only 10–15% have participated in large-scale AI projects. This is a modest figure regionally, but the growth potential is enormous.
The key weakness lies in real-world deployment experience – particularly in transitioning models from lab settings to real-life environments with noisy data, limited resources, and stringent security requirements.
At many tech firms in Vietnam, including ours, we’ve observed that when given real project exposure, Vietnamese engineers adapt extremely quickly. Some even master complex architectures like transformers or Vietnamese-optimised computer vision models.
In your field of AI, what are the most common technical challenges your team encounters?
One of the biggest challenges FSI faces in document digitisation projects is handling unstructured Vietnamese data – especially outdated administrative documents with poor scan quality, a mixture of printed and handwritten text, official stamps, non-standard tables, and obsolete Vietnamese fonts.
For instance, in digitisation projects with government agencies, FSI had to develop extraction models combining computer vision, natural language processing (NLP), and post-processing error detection algorithms. These models not only needed high accuracy but also had to process tens of millions of pages quickly, all while complying with strict national data security regulations.
See also: How Hanoi’s Public Service Centre adopted D-IONE for their digital transformation journey.
Another hurdle is the high variability in Vietnamese data. A single type of document might appear in 5–7 different layouts depending on the locality. Standardising input and training models that can “tolerate deviation” is a highly complex task.
Is there a difference in problem-solving approaches or AI mindsets between Vietnamese engineers and their international counterparts?
Yes, and it’s understandable. International engineers – particularly from the US or Japan – tend to be methodically trained, highly process-driven, and product-oriented, always focusing on end-user experience.
Vietnamese engineers, on the other hand, are more flexible and resourceful in constrained environments. They often start with concrete problems rather than system-wide design – this is both a strength and a limitation when aiming for global competitiveness.
In international collaborative projects I’ve worked on, we typically formed mixed teams: Vietnamese engineers handled data and logic layers, while foreign engineers focused on system integration. This combination proved highly effective.
What are the advantages and limitations in accessing cutting-edge AI technologies and models in Vietnam? How do you stay up to date?
In theory, access to knowledge is no longer a barrier. Platforms like arXiv, GitHub, Papers with Code, and HuggingFace are freely available. The real bottleneck lies in the internal capability to experiment and implement these new models in real environments.
Many advanced models today require powerful GPUs – for example, LLaMA 3 requires A100 or H100 – yet computing infrastructure in Vietnam is still limited. Additionally, the absence of AI sandboxes or secure environments for tech trials slows down adoption.
At FSI, we maintain a dedicated R&D team that monitors international AI publications weekly, combines that with specialist forums and internal experimentation – it’s how we keep pace with the sector’s breakneck evolution.
Apart from technical skills, what is the most important ability for a Vietnamese AI engineer to compete globally?
If I had to pick one, it would be product mindset. Many engineers are great at coding and modelling but can’t answer questions like: “How will users interact with this feature?” or “What is the inference cost per run, and can it be optimised?”
Bridging the gap between technical expertise and practical application is what separates good engineers from influential ones.
Soft skills are also essential – especially communicating technical concepts in English and the ability to work across cultures. These are non-negotiable in global environments.
Does the Vietnamese work environment support AI research and development?
Compared to five years ago, the landscape has improved significantly. More venture capital, corporate AI centres, and institutional support have emerged. But challenges remain: data accessibility is difficult, IP protection laws for AI are unclear, and high-risk R&D lacks policy support.
Vietnam urgently needs a legal framework for emerging technology experimentation (AI sandbox) and incentives for industry-university collaboration to advance applied research.
A positive trend is that tech companies like FSI have started building in-house R&D centres, partnering with universities, and deploying test solutions in real-world projects – not just for business value, but to serve as “innovation nurseries” for Vietnamese AI.
How should Vietnamese companies invest in R&D to create globally competitive breakthrough products?
The fundamental condition is to see R&D not as a cost but as a strategic investment. And sustainable R&D cannot be done in isolation – it requires collaboration between businesses, academia, and government.
At FSI, we’ve committed to long-term projects like AI models for administrative form extraction, Vietnamese text with diacritics, and end-to-end data assurance. One project took 18 months to build a model that handled just 0.5% of extremely unreadable data – yet that portion helped our system meet the strict requirements of a national data initiative.
What makes you most proud as an AI engineer, and what advice would you give to the next generation?
I’m proudest when the models we build aren’t just algorithms – but actually used in real life. Even modest impacts like helping a civil servant process records faster, saving a business storage costs, or sparing a citizen from hunting down lost documents – these give our work real meaning.
My advice: AI isn’t magic – it’s a tool. But if you use it at the right time, in the right way, and with a clear purpose, it can drive immense change. Master the fundamentals, understand the problems deeply, and always ask: “What value does my solution bring to the user?”
You may also be interested in:
