Learning by talking: Instead of fishing for answers, students explain concepts to bot

Sign up now: Get tips on how to grow your career and money

(From left) Singapore University of Technology and Design (SUTD) alumnus Akshaya Rajesh, GenAI engineer, and Dr Sumbul Khan, senior lecturer in SUTD's Science, Mathematics, and Technology cluster, on Oct 28, 2025.

Singapore University of Technology and Design alumnus Akshaya Rajesh (left), GenAI engineer, and Dr Sumbul Khan, senior lecturer in SUTD's Science, Mathematics, and Technology cluster.

ST PHOTO: AZMI ATHNI

Follow topic:

SINGAPORE – At the height of the Covid-19 pandemic, social distancing rules meant that Ms Akshaya Rajesh could not hold study sessions with her friends.

She ended up explaining what she learnt out loud to a stuffed toy bear in her hostel room at the Singapore University of Technology and Design (SUTD). “I remember thinking, ‘If only this bear could talk back to me and ask questions,’” said the 25-year-old generative artificial intelligence (AI) engineer, who graduated in 2022.

Recent advances in conversational chatbots have given her the tools to create the Feynman bot – named after a learning technique developed by Nobel laureate Richard Feynman. The physicist believed that to truly understand a topic, one should be able to explain it to oneself or others.

Unlike most popular AI chatbots used by students to get instant answers, the Feynman bot flips the script by testing the user’s understanding, asking questions like a friend would.

For more than a year, Ms Rajesh has used her free time after work – about four times a week – to develop the bot. To test it, she approached Dr Sumbul Khan, a senior lecturer at SUTD’s Science, Mathematics and Technology cluster, for help in letting a group of 70 first-year students use the bot as part of a six-week maths bridging course.

“We found that students who used the bot reported greater confidence in their knowledge during biweekly quizzes, as opposed to those who did not,” said Dr Khan.

“If somebody just gives you the answers, that is passive learning. But when you struggle with the concept a bit, that helps you in remembering the concepts better.”

For the bot to ask questions, students first need to upload their study materials such as lecture notes. It can simulate anyone – from a primary school pupil to a university graduate.

“The limit is around three to four times that users can say they don’t know the answer, then the bot will suggest an approach and give a new question based on this weakness,” said Ms Rajesh.

She used OpenAI’s large-language model to build the bot. This helps it tailor questions to the user’s level of study and reduces the chances of hallucinations, said Ms Rajesh. Hallucinations refer to inaccurate and fabricated responses made up by AI models.

“Thermodynamics can be explained in five different ways, and the way a kid and a university student learn about it is very different,” she added.

First-year SUTD student Shrinidhi Sivakumar found the bot useful for understanding the fundamentals of advanced topics in physics.

“When I learnt about projectile motion recently, there were so many formulas and I didn’t know when to use each one,” said the 19-year-old, whose last encounter with physics was in secondary school. “Everything seems new to me after entering university, and I can’t really expect my professor to explain everything from the basics. So when I’m alone, the bot helps to make sure that my understanding is correct.”

Fellow first-year student Kumaravelu Kishore used the bot thrice a week while studying for his recent mid-terms, and liked how it clearly outlines learning objectives for him to complete. But the 22-year-old said he will likely continue doing practice papers.

“The best mode of studying would be a combination of both.”

One improvement both students hope to see is the ability to not only generate questions but also give answers illustrated with diagrams and graphs.

From January to September 2026, a trial involving 350 Secondary 4 students from Raffles Girls’ School will allow the team to collect more feedback and data. The students will be using the bot at home while doing homework, and also during biology lessons.

If the bot becomes widely used in classrooms, another possible upgrade would be to aggregate the strengths and weaknesses of students in each class, which will help teachers in their lesson planning, said Ms Rajesh.

“The goal is to bring it to kids that don’t have access to tuition or have to study by themselves. They, of all people, need access to tools that are accurate, user-friendly and cheap.”

The Feynman bot can be accessed

here

.

The bot will continue to test the user’s understanding of a topic, until a set of learning objectives is completed.

PHOTO: SCREENSHOT FROM FEYNMAN BOT

See more on