NTU Singapore saves 14 days of admin work monthly with GenAI bot

Nanyang Technological University, Singapore (NTU Singapore) has collaborated with AI solution provider CloudMile to upgrade its Lyon Bot student assistant with generative artificial intelligence technology, resulting in substantial reductions in administrative workload.

The enhanced Lyon Bot integrates Google's Gemini large language model, enabling more responsive and accurate handling of student queries, particularly in the area of accommodation, and has led to savings of over 14 days of manual work each month for the university's staff.

Hybrid approach to student inquiries

The transition to generative AI marks a shift in NTU's management of student housing enquiries, moving from the university's previous intent-based Natural Language Processing (NLP) chatbot system. Mr Ng Kee Haur, Deputy Director, Enterprise IT, Centre for IT Services at NTU, outlined the nature and benefits of this advance.

"We are transitioning from an intent-based Natural Language Processing chatbot to a GenAI chatbot for student housing inquiries to enhance flexibility, scalability, and user experience," said Mr Ng Kee Haur. "While GenAI can handle complex, unstructured queries, there is still a role for the intent-based chatbot in specific scenarios. This hybrid approach allows us to balance efficiency with precision."

The original Lyon Bot, activated in 2020 to assist around 6,000 new students with university enrolment, was based on Google Dialogflow Essentials. Although effective in handling routine queries, the system struggled to adapt to evolving student expectations and required significant manual input to update and tag data.

AI-powered process improvements

With CloudMile's assistance, NTU revised Lyon Bot's architecture, introducing generative AI features that include multi-layered processing methods, encompassing query comprehension, demand classification, and real-time information retrieval from various centralised data repositories. According to the university, this has led to a notable increase in the accuracy and speed of conversational responses, as well as a decrease in error rates.

Alvin Ong, NTU's Chief Information Officer, described the impact of these efficiencies:

"With the new AI-powered chatbot, we are currently saving around 14.5 days' worth of work every month. This allows our staff to focus on tasks that require human insight and creativity."

The university notes that the new system enables staff to dedicate more time to roles that require human judgment and creative problem-solving, rather than routine administrative tasks associated with student inquiries.

CloudMile's role and expertise

CloudMile supported NTU through consulting and technical integration services that were tailored to the educational environment's requirements. The company brought together expertise in generative AI applications, data governance, and large-scale international project deployment to support the university's objectives.

Jeremy Heng, Managing Director, Southeast Asia at CloudMile, commented on the collaboration's significance:

"The transformation of NTU's Lyon Bot is a testament to our commitment to enhancing human-computer interaction through AI. CloudMile will continue supporting institutions and enterprises in leveraging AI for innovation excellence."

CloudMile's involvement included designing a scalable framework for AI-based conversational support, ensuring the solution could accommodate the evolving demands of a large educational institution, both technically and operationally.

Long-term efficiency and engagement

NTU reports that the generative AI-powered Lyon Bot has improved overall engagement among students, making use of the virtual assistant for their enquiries. The automated approach is also considered to provide scalable support for future enrolment cycles and additional administrative purposes.

The university continues to emphasise the dual utility of both generative AI and intent-based chatbot systems, acknowledging that the combination provides the flexibility to address both structured and open-ended queries with appropriate levels of precision and efficiency.