9X medical guy "scoring" international AI, teaching machines to understand Vietnamese people
(Dan Tri) - From a medical data researcher in Japan, Nguyen Minh Phat turned to AI after the Covid-19 pandemic. Currently, he participates in testing international AI models, helping machines understand Vietnamese people more naturally.
"Why do I still have to hand-copy thousands of test tubes like this?"
Amid the most stressful days of the Covid-19 pandemic, that question was constantly on Mr. Nguyen Minh Phat's mind. A few months later, he built a system to automatically print barcodes. A few years later, that same thought led him to a job evaluating global AI models.
From the biomedical lab to a journey with AI
Immediately after graduating with a Master's degree in Medical Sciences from Kanazawa University (Japan), Mr. Nguyen Minh Phat, born in 1993, returned to Vietnam to participate in the fight against the epidemic on the front lines.
At that time, he still thought his path would be linked to testing, medical data and health care problems.
In 2021, during the peak of the Covid-19 epidemic, testing sites had to take samples from thousands of people every day.
According to normal procedures, medical staff must hand-write information on each test tube before taking samples. As the number of samples skyrocketed, that seemingly small task became a bottleneck that was both slow and confusing.
"At that time, the medical team had to handle a very large amount of samples. Hand-copying information for thousands of test tubes became a silent bottleneck but caused great pressure," Mr. Phat recounted.
In the midst of a stressful job, he constantly wondered: "Why hasn't such a repetitive job been automated yet?".
From that question, he came up with the idea of building a system that pre-prints information with barcodes on each test tube, replacing hand-written notes. When the system goes into operation, the workload is significantly reduced, while limiting the risk of sample confusion. This solution then entered the Top 50 of the HIS-COVID 2021 competition.
That's when he realized that the power of technology lies in removing the exact knot people are stuck in, not in showing off.
"If a small automation step has liberated hands, then AI is the lever to liberate thinking capacity on a large scale," he shared.
Switching to AI, for him, is not about leaving the mission of taking care of people, but he sees it as a way to go deeper into the roots: helping people increase productivity, create more value and thereby have the conditions to build better systems for the community.
"I want Vietnamese people to proactively steer the wave of technology, instead of being swept away by it," he said.
With Vietnamese tasks, the difficult part is not just to see whether the AI uses the right or wrong words, but to evaluate whether the answer is close to how Vietnamese people actually ask, ask for help, refuse, give suggestions or express emotions in life.
AI just being right is not enough
For many people, working in the field of AI is often associated with images of engineers writing algorithms, training models or building giant computing systems. But Phat's job is different.
Every day, he receives tasks on a dedicated labeling tool. He is the one who sets the tasks for AI through highly personalized prompts, simulating real needs that users may make in life.
After the AI model responded, he continued to perform A/B scoring between different answer versions.
This job is not simply choosing the "right" or "wrong" answer, but comparing which option is more useful, more natural, safer, and exploiting contextual information in the simulation scenario more appropriately.
"I have to review every word to make sure the AI not only answers correctly but is also truly useful," said Mr. Phat.
For example, if a person asks about an upcoming trip, a good AI assistant doesn't just give a list of famous places. It needs to understand where the user has gone, what type of experience they prefer, or what constraints they have on their schedule.
"It's a shift from a lookup machine to a real assistant and secretary," Mr. Phat said.
However, to get to that point, AI models still need many layers of human review. During the work process, Mr. Phat must read each feedback carefully to see if the AI is following the right data, is it over-interpreting, is it clumsily using personal data, or is it touching the user's sense of privacy.
He emphasized that people doing this work do not directly train the model in the core technical sense. They are like a link in the quality control chain: detecting errors, noting what's wrong, and passing that feedback on to relevant departments.
For Mr. Phat, the work of "grading" AI shows an interesting thing: even in the most modern technology systems, the role of humans is not lost. It just shifts from directly replacing machines to inspecting, navigating, and helping machines understand humans better.
AI is right in meaning but wrong in “language sense”
According to Mr. Phat, the biggest challenge when evaluating Vietnamese AI lies not only in grammar or information accuracy, but in language nuances.
Vietnamese is complicated by the system of pronouns, speaking according to roles, regions and communication contexts.
An answer may be technically correct, but if it's too polite with elders, too cold in a situation that requires sympathy, or uses the wrong locale in the wrong context, it's still considered "out of tune."
"AI can be 90% correct in terms of semantics and vocabulary but still fails if the language sense is wrong," he said.
As an example, he asked AI to write a birthday wish for his mother. Instead of creating a warm wish from a child, the AI returns a dry text in the style of health advice, reminding to go for regular checkups, eat light meals, and monitor blood pressure.
"AI is not wrong if viewed from the user's occupational data, but it is wrong in context. It assumes that person is always a doctor, forgetting that in that moment, he was just a son who wanted to send his mother a word of love," Mr. Phat analyzed.
According to him, such errors cannot be detected just by looking at grammar. Evaluators must put themselves in the user's position, understand the context, habits and unique nuances of Vietnamese communication culture.
“We not only check whether the AI is right or wrong, but also whether it makes the user feel understood,” he shared.
In addition to being a collaborative expert in a global AI project, Mr. Phat also supports startup groups and local businesses in applying AI to production and business activities.
For him, these two jobs are not separate but complement each other: one helps him access the latest AI trends, the other is an opportunity to bring that experience back to serve local businesses and people.
In addition, Mr. Phat also developed his own AI system called Tin (The Integrated Network), built so users can control an "AI team" right on their phone.
He is currently developing this system in the testing phase, focusing on allowing users to assign tasks by voice, coordinate multiple AI tasks at the same time, and operate mainly on the phone instead of depending on the computer.
Looking to the future, he believes that AI will continue to change the way people work, but that does not mean that humans will be replaced.
"Anyone can rent an engine, but where to install it to create value is important. Young people should not just learn how to use AI faster, but need to understand a field deeply enough to know how to assign tasks, monitor and approve the results of AI," he shared.
He believes that Vietnam's advantage lies in its young force, which is quick to absorb and understand local culture and needs. If you know how to combine AI coordination skills with expertise in a specific industry, young people can create suitable solutions for domestic life and production.
For Mr. Phat, his journey can be summed up in a simple sentence but it is also something he always pursues:
"I am a person who does many things around AI: during the day I teach about AI, at night I teach AI about people."