Unlocking the vault: NYP trio uses AI to ‘immortalise’ part of Singapore history

SINGAPORE – On the 10th day of the fourth lunar month in 1906, a group of Singapore’s pioneers gathered for a general meeting. On their agenda: Responding to a British government investigation into shipping interests, establishing regional schools, and setting fund-raising rules that awarded donors of 1,000 yuan a portrait in a central hall.

These historical details, including thousands of pages of meeting minutes dating back to 1906 and countless black-and-white photographs of faces, were meticulously recorded in handwritten traditional Chinese by the Singapore Chinese Chamber of Commerce and Industry (SCCCI). But they remained largely unseen by the public for over a century.

While the archives were never at risk of being lost, the information within them was effectively locked away in archaic scripts that few modern Singaporeans could read.

This changed in May 2025 when a trio of students from Nanyang Polytechnic (NYP) – Mr Prakhar Trivedi, 20, Mr Joon Jun Han, 20, and Ms Toh Zheng Yu, 23 – took up the challenge of digitising these artefacts, using artificial intelligence to make these “hidden” records more readable and accessible to the public. 

They were asked to help digitise up to 2,000 artefacts from the SCCCI for ease of storage, organisation and public showcase, and the team rose to the challenge. 

They co-founded ArchAIve, an AI-powered tool that digitises handwritten Chinese documents and calligraphy, helping to preserve Chinese cultural heritage in Singapore. 

It uses optical character recognition technology to process images of old scripts, and leverages Alibaba’s “Qwen” visual language models. 

Unlike standard tools that struggle with the irregular spacing of ancient calligraphy, this system is optimised to recognise four major styles: clerical, standard, semi-cursive and seal scripts. 

It is able to translate handwritten traditional characters into modern simplified Chinese and English, detect historical figures using facial recognition, and provide concise summaries.

Mr Trivedi said all he needs is an image of the document. “You scan the meeting minutes and you have an image, and you just feed it into the AI,” he added. 

The system automatically detects the document type and engages relevant models to perform the transcription, navigating the “backward” reading order of historical Chinese and handling faded or stained backgrounds with an accuracy rate of up to 95 per cent.

For the chamber’s vast collection of historical photographs, the team employs facial recognition algorithms to bridge the gap between faces and identities, Mr Trivedi said. 

The AI detects individual faces and analyses specific features to create a unique digital profile, he added. While the initial identification of a figure requires a researcher to manually key in a name, the AI subsequently “learns” that face. 

Ms Toh said the system even accounts for modern challenges like face masks from the Covid-19 years, using a “merge” feature to ensure different photos of the same person are correctly linked. 

Mr Joon said: “You click one button, and ‘poof’ – it is like a magic wand.”

Despite the AI’s speed, however, human vetting is still required to review the AI’s output side by side with the original scan. 

“This is part of our human-in-the-loop approach,” Mr Joon said. “AI is just maths at the end of the day, we can’t fully trust it.”

But even with the human vetting, which will bring the total time spent digitising one artefact to about five to seven minutes, it still saves overall time on the manual work a heritage researcher would have had to do in the past. 

This efficiency has allowed the team to digitise close to 400 artefacts within the first few weeks of ArchAIve’s pilot deployment.

The accuracy rate for photos ranges from 95 per cent to 99 per cent, said Mr Trivedi, and for meeting minutes, it is about 85 per cent on average. 

And for better written scripts, it can go up to 95 per cent, although scripts with faded and cursive handwriting will get slightly less accurate results. 

Some photos and meeting minutes are currently available on ArchAIve’s website, https://archaiveapp.com/. It includes an interactive element for the public – a chatbot that allows users to interact directly with the digitised knowledge.

Unlike typical AI that might “hallucinate” – generating false information – this chatbot draws directly from verified metadata. Users can ask specific questions, such as “What did the SCCCI do for education in 1906?”, and receive factual answers based on the scanned documents.

Mr Joon said the “online gallery” will be easily accessible to the younger generations on their phones – it is more interactive and would help them understand history better compared with visiting a museum and just staring at text.

Ms Tan Mei Hui, senior executive at SCCCI, said the goal is to move beyond simple electronic storage, and potentially have these artefacts for public showcase. 

“More than preservation challenges... the challenge of digitising Chinese records lies mainly in enabling and maximising readability, searchability and ease of editing of the information in the records,” she added. 

By utilising ArchAIve’s platform, this works towards ensuring that the “Singapore story” told from the chamber’s perspective is accessible to all. 

“We hope to work towards an online collection of selected artefacts for public showcase, as well as better documentation and management of our artefact records,” Ms Tan said.

For the students, the project – which is expected to conclude its current phase in June – became a rare bridge to the past. Mr Joon, who had always dealt with simplified Chinese in school, found it interesting to manually read the traditional characters.

“Previously, I wouldn’t understand that... it definitely enhanced the depth (of my knowledge),” he said, adding that his biggest takeaway was learning to read “fan ti” (traditional script).

Ms Toh, who admittedly struggled to read traditional characters initially, found herself matching strokes and characters to understand the meaning.

Mr Trivedi, who is Indian, said he gained a deeper appreciation of Chinese history, and it made him “prouder to be Singaporean”, because of how easy it is to feel out of touch with what Singapore’s forefathers and other historians did to bring the country together.

“I can access this Chinese culture,” he said, as historical texts become more accessible with English translation. “I can learn from it; I can understand how people thought, what they did, their practices, all these different things.”

“I think what this really reflects… is the power of AI technology to democratise access to culture and also to bring people together across cultural boundaries,” Mr Trivedi noted. “I think this is going to be a great way to bring Singapore together, and it is especially instrumental for us because we are a cultural melting pot.”