You are here


Unlocking the potential of artificial intelligence

Bifrost's platform gives artificial intelligence engineers access to affordable and high quality data quickly.

A synthetic street scene extracted from a game engine used for video games.

AUTONOMOUS vehicles, robots and drones that use artificial intelligence (AI) require hundreds of thousands of images and videos to learn about the world around them. Collecting these images is a time-consuming and expensive step when building AI applications. Instead of capturing data from the real world, tech startup Bifrost has created an innovative platform that is able to generate custom virtual worlds for this purpose.

"The biggest bottleneck for AI development is data. There's not enough of it, and it's expensive to get and label," said Charles Wong, CEO of Bifrost. He noted that to train an AI model to recognise objects on the street, it would have to be shown hundreds of thousands of pictures of streets. These images have to be varied, and each object has to be manually labelled with descriptions like "this is a car" or "this is a traffic light".

To address this issue, the company developed the Bifrost Synthetic Data Engine, which uses data that is artificially created by utilising the same graphics technology used to generate virtual environments in movies and video games.

"Our engine automatically generates a realistic looking virtual world, where we can simulate rare and dangerous occurrences like a building collapsing, or a car accident. These worlds let our clients capture rich data at a fraction of the time and cost," said Mr Wong.

Market voices on:

The synthetic data platform will give engineers access to affordable, fast and high quality labelled data. "Our process reduces AI development time to days instead of months or years. Not only do we speed up existing AI teams, we also enable companies who have no data or AI capabilities to get started on their AI journey," explained Mr Wong.

As AI engineers themselves, Mr Wong and his co-founder, chief technology officer Aravind Kandiah, came upon the idea for Bifrost after personally experiencing the pains of data collection and labelling. "A company that Aravind worked for had hired 10 interns to sit around and label data all day. This felt like an archaic way of developing AI, and we knew there had to be a better way."

The partners began developing an early prototype of the solution for their Bachelor's degree thesis earlier this year. They started running Bifrost full time after graduating from the Singapore University of Technology and Design this year.

"It was always going to be a challenge as a new startup with young faces trying to convince customers that our services and tools actually work. I'm very grateful that our current and future customers are giving us the chance to prove ourselves with what we believe is potentially disruptive technology. Techblazer also serves as an excellent source of validation for us," said Mr Wong.

The company has already secured a number of customers from sectors like mobility, logistics and video analytics. "We have a greater vision for the product and its future roadmap, so development work is never done."


WHEN School of Science And Technology (SST) student Carl Ian Voller took part in the Australian Informatics Olympiad in 2017, he noticed during the training sessions that each time students were given a problem to solve that would be judged by the class afterwards, they would have to repeatedly disconnect, then reconnect to the display to present their code.

That inefficiency sparked an idea in Mr Voller to simplify the process by making every student's code easily accessible to multiple users in a directory. Any changes to the code made by a student or teacher would also be updated on both their devices.

This idea eventually led Mr Voller and his co-founder, fellow SST student Qin Guan to create CodeCollab, an online collaborative code editor and compiler that allows its users to test and run code in real-time.

CodeCollab was launched in late 2018 and is currently in its beta release; with the next release scheduled for December 2019.

The solution is also being incorporated into the O levels Computing Curriculum of SST, as well as into the A levels Computing Curriculum of Dunman High School.

Said Mr Voller: "Our next step is to reach out to even more schools that offer computing to find out how they can integrate CodeCollab into their curriculum."


Developed by a five-person student team from Anglo-Chinese School (Independent), MedACS is a pill dispensing device and a smartphone app that helps patients and their caregivers better manage medicine taking.

The solution comprises a device that stores the medicine and will dispense the required dosage at scheduled times, as well as provide reminders using sound and light. It will also alert the caregiver through the app when the medicine has been taken.

The team has developed a fully-working prototype, and the students plan to the refine the product after their examinations.

"MedACS helps patients and caregivers by automating medicine dosing and allowing remote care. The Techblazer Awards has been a great platform for us to learn from industry leaders and apply our engineering and scientific knowledge to create practical solutions," said Dylan Tay, one of the students behind MedACS.