Singapore-based AI startup Pand.ai Pte. Ltd. today announces the official launch of its latest AI engine, TE01, which is able to generate Natural Language Processing (NLP) models that promise to cut error rate for intent classification, an important parameter to measure the model's accuracy, by up to 25% while also significantly slow down NLP decay.
Pand.ai is one of the first companies in the region to deploy Transfer-Learning based NLP chatbots commercially. Last year, it won a competitive open tender to build an AI chatbot for Great Eastern Life Singapore (GELS) on the strength of TE01, a prototype at that time. Since then, models generated using TE01 had been put into production, starting with the launch of GERICA, the first integrated AI chatbot offered by GELS for its agents to perform such inquiries like claim status check and policy details.
Commenting on the application of TE01 model for Great Eastern's chatbot GERICA, Mr Tan Meng Siong, Senior Vice President of Digital Transformation at Great Eastern said, "A key objective for us was to build the best chatbot for our financial representatives and customers and to take financial advisory to the next level, thus we wanted to work with a deep-tech solution provider that has the best technology and the most experienced in the insurance domain. We were impressed by Pand.ai's state-of-the-art TE01 engine. Since GERICA made its debut in April, we have seen very high satisfactory score and deep engagement between the chatbot and our pilot group of users, with some financial representatives asking GERICA hundreds of questions in a week."
The new TE01 engine is built based on Transfer-Learning, the state-of-the-art NLP method, which was first published by the research community in mid-2018. Transfer Learning-based NLP models, such as those generated by TE01, are more elegant solutions compared to existing NLP methods, which are often shallow approaches that trade expressivity for efficiency.
"We are incredibly excited about the launch of TE01. Our internal benchmark shows given the same training data set, TE01 produces NLP models that outperform previously state-of-the-art methods by reducing the error rate by more than 20% consistently. This allows us to create AI chatbots for our clients with a higher degree of accuracy and in a shorter amount of time," said Chuang Shin Wee, co-founder and CEO of Pand.ai.
The development of TE01 was led by Pand.ai's co-founder and Chief Data Scientist, David Low, who is also an adjunct lecturer in Machine Learning at the National University of Singapore.