Project KAIVA

• Main Author: Pua, Ivan Jun Hao
• Other Authors: Vun Chan Hua, Nicholas
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Software::Software engineering
• Online Access: https://hdl.handle.net/10356/171983
• Institution: Nanyang Technological University

Artificial Intelligence (AI) with generative capabilities has become a topic of extensive discussion in recent years. Products driven by Generative Artificial Intelligence (GAI), such as ChatGPT, have gained immense popularity since their launch. As a result, the broader audience has witnessed the remarkable potential that GAI holds. This project is developed in collaboration with Kuok Singapore Limited (KSL). It is a Proof of Concept (POC) with a goal to integrate cutting-edge GAI capabilities which will enable employees to access crucial organizational information and boost workplace efficiency by streamlining repetitive daily routines. Project KAIVA – acronym for Kuok Singapore Limited Artificial Intelligent Virtual Assistant introduces an AI Virtual Assistant to be integrated with the Microsoft (MS) Teams platform. Its primary objective is to streamline employees’ daily work processes by providing a centralized point of initiation, as well as enabling employees to access important organizational information. The software automation and integration with MS Teams will be facilitated through Workato, a prominent Integration Platform as a Service (iPaaS) company proficient in workflow process automation. Meanwhile, the generative capabilities of KAIVA will be facilitated by OpenAI, the developer most renowned for the development of ChatGPT. The process of generating responses and the underlying system architecture will be discussed further in the subsequent sections of this report. The virtual assistant utilizes a technique which enables domain-specific text generation using vector embeddings, eliminating the necessity for training a dedicated Large Language Model (LLM). Virtual assistants or chatbots of such characteristics may have the potential to revolutionize the conventional approach to developing virtual assistants where a rule-based or retrieval-based approach is adopted. In contrast to the conventional approaches, GAI provides the virtual assistant with a heightened level of dynamism, resulting in responses that are more dynamic and thus enriches the interactions with its users. However, each technique brings along its own set of strengths and limitations, which will be examined further in subsequent sections of this report.