TutorMe

Inspiration

In Singapore, 26.65% of families are considered low income families, with an average monthly income of just $1.9k. While these families are able to provide the basic needs for their children, many of them are unable to afford quality education, which is essential for breaking out of the poverty cycle.

For example, the average price of tuition is $112 a month, which is unaffordable for low income families. As a result, children which face difficulties in their studies are often unable to seek the required guidance to improve.

Education is known as "the great equalizer", but not all children start on an equal footing. Many families lack the financial means to provide additional educational support for their children, so this disparity in resources ends up limiting opportunities for social mobility through education, demotivating both the parents and the students.

Our group hopes to address this gap, providing unhindered access to quality tutoring through the latest technologies regardless of the social conditions of children, and uplift both parents and students.

What it does

AI Assistant for students, tutoring them in subjects taught in school. Main features include

Free access to AI Tutor to aid students to better understand school work
Specialized chatbots which have been tuned to various subjects
Image-based queries, allowing for AI Tutor to help students which do not understand topics well
Voice prompting, allowing users to seek help from the AI Tutor through speech
Inbuilt verification of responses through Chain of Verification (COVe)

How we built it

Built using React and Next.js, our solution integrated free APIs from Google AI Studio on State-of-the-Art Gemini models.

With regards to the LLM, we fine tuned it on a portion of the summary notes and exam practice papers curated by Singaporean students such that it would be able to answer the various queries students may have. We also included chain of thought and chain of verification, ensuring high quality in responses.

Challenges we ran into

There were many setbacks faced during the development of the project. A significant setback being the availability of data to train on, as we require the model to be trained on Singapore’s context. However, after extensive research, we managed to find summary notes and exam practice papers curated by a group of students in Singapore called the “Holy Grail” to use as training data to fine tune our LLM model with.

Accomplishments that we're proud of

An accomplishment that we are proud of would be integrating our group’s understanding of various state-of-the-art models to develop this solution. Individually, our team members bring diverse experiences in building a range of deep learning models, including foundational large language models (LLMs) like ChatGPT and Google Gemini, as well as advanced multi-modal models and vision-language models (VLMs).

By leveraging this collective expertise, we were able to create a solution that combines the strengths of these cutting-edge technologies.

What we learned

We learnt how to maximize the capabilities of LLMs to make a difference in people’s lives, especially those in need and disadvantaged.

We also learnt the importance of prioritizing user experience through intuitive user interface designs.

What's next for TutorMe

One feature we are keen to implement is a system for tutors to vet the LLM responses such that we would be able to further improve the LLM and ensure that the responses provided would be kept to a high standard