Remember Dory from “Finding Nemo”? The lovable blue fish who is forgetful, easily distracted, and often off in her own world?
While her quirks are endearing, they also mirror what many individuals with attention-deficit/hyperactivity disorder (ADHD) experience in real life.
ADHD affects about 3.1% of adults worldwide, and in the US alone, one in nine children between the ages of three and 17 has been diagnosed with it.
Some people with ADHD, like Bryan Chee from Singapore, report trouble with time management. Recognising their struggles, one statistics student from China developed a time management app to help.
Besides being an avid statistics student, Tao is also really passionate about playing the piano. Source: Haitao Tao
From numbers to AI
Growing up, Haitao Tao has always been drawn to numbers. After finishing his A-Levels at Adcote School Shanghai, he packed his bags for the London School of Economics and Political Science’s (LSE) BSc Mathematics, Statistics and Business.
“I wanted to blend statistics with finance and get into something like quantitative trading,” he says.
But once he settled into LSE, things changed. Tao realised he wasn’t as passionate about finance and accounting as he thought. So, in his final year, he switched to artificial intelligence — specifically, machine learning and deep learning.
If that sounds like a big leap, Tao insists it’s not.
“AI is actually just statistics in action,” he explains. “For example, if you look at large language models (LLM), they’re simply predicting the next token, which is the next word. It’s all about statistics and probability at its core.”
So, what does learning AI look like?
At LSE, it starts with theory. “My deep learning teacher was great,” Tao shares. “He integrated many state-of-the-art topics into the data course.”
For example, when DeepSeek was trending, he added it to the lecture and broke down the programme’s technical aspects.
Still, with AI evolving so quickly, some wonder if a degree is even necessary. For Tao, the answer is clear: yes.
“I think having a strong foundation is crucial,” he says. “If you don’t understand the principles, the techniques, and the technology behind AI, it’s difficult to keep up with the trends or grasp the concepts driving new advancements.”
When he was not making a difference using statistics, Tao travelled to the Netherlands and Spain. Source: Haitao Tao
Languages coders need to speak
Throughout his degree, Tao tried different programming languages, like JavaScript and TypeScript for web applications.
JavaScript is a popular language used to make websites interactive, like adding animations, buttons, or forms that respond to users.
TypeScript builds on JavaScript by adding extra features like error checking and more structured code. This helps developers catch mistakes early and makes their code easier to manage.
Tao also worked with R, an open-source statistical programming language specifically designed for statistics students to conduct data science and analyse large datasets.
But when it comes to the must-have tools for statistics students diving into AI and machine learning, Tao has two clear favourites: Python and PyTorch.
Python is a high-level programming language often used for tasks like web development, data analysis, software development, and automation.
PyTorch is a software-based open-source deep learning framework that builds on that foundation.
“PyTorch is one of the libraries in Python,” Tao explains. “It’s made specifically for deep learning. So you write your code in Python, and PyTorch gives you the tools to build neural networks. For statistics students looking to break into AI, it’s incredibly useful.”
Most to-do apps use text to show tasks. EasyDDL takes a new approach by using visuals to help you understand and manage your time better. Source: Haitao Tao
An app that visualises time
For Tao, the best way to learn a programming language is by doing. Instead of just reading tech notes or copying code examples, he prefers to jump into a project and figure things out along the way.
That’s how he got started with his time management app, EasyDDL.
“At the time, I was still figuring out JavaScript and TypeScript,” he says. “I didn’t know everything about web or app development, but I wanted to challenge myself.”
He started by sketching the design by hand, and now, that idea has turned into a real app available on the Apple App Store.
“Developing the app gave me a way to apply those skills while continuing to learn through hands-on experience.”
Tao had a personal reason for building the app too. “I’ve always found it very hard to manage and visualise the time,” he says.
Traditional to-do lists didn’t help. “You can only write your tasks; they don’t give you a clear visual of how time is actually used. I built EasyDDL to help me see and manage time better, and I hope it can be useful to others facing the same issue.”
While EasyDDL focused on software development, Tao wanted to work on something more closely related to AI and deep learning. That led to his second project: a multilingual restaurant review analyser.
Which, disclaimer, didn’t hurt that he could also use it for a school assignment and submit it to a hackathon.
Tao presenting his restaurant review analyser in class. Source: Haitao Tao
No more fights over which restaurant to dine at
The idea came when Tao and his teammates were brainstorming for a hackathon. They got hungry halfway and decided that was all the inspiration they needed to make a restaurant recommender.
“Most people use Google Maps to find places to eat, but it just gives you a list,” Tao says. “It’s not tailored to what you like or need.”
To bring their idea to life, the team turned to Hugging Face, a Python library that provides access to thousands of pre-trained Transformers models for natural language processing (NLP), computer vision, audio tasks, and more.
There, they picked one of the restaurant review datasets available and started fine-tuning a model using what they’d learned at LSE.
After experimenting with different methods to optimise it, they uploaded their final version to Hugging Face so others could download and try it themselves.
But the process wasn’t without challenges. “The biggest challenge was that the dataset was in Dutch,” Tao says. “We spent a lot of time testing different pre-trained models to figure out how to work with it.”
Then there was the cost. “We had to pay for training time on Google Cloud and upgrade to a better GPU just to get the model to run properly,” he adds.
Even with the long hours and added expenses, Tao says the experience was worth every bit of effort.
“It was the first time we took what we learned in class and turned it into something real,” he says.
“We didn’t just build the tool, we also learned how to collaborate effectively using platforms like GitHub, and how to manage the entire development process from start to finish.”
