This thesis investigates the potential of an AI-powered voice assistant in simplifying and securing financial transactions in a Peer-to-Peer (P2P) payments application, integrating cutting-edge Machine Learning technologies on-device such as BERT for text classification and a custom Dialogue State Tracking mechanism. Through a meticulous design and development process, the study explores the integration of voice commands into a real iOS app to facilitate in-app operations, prioritizing user privacy through local data processing, and enhancing user interaction and accessibility via an intuitive UI/UX design leveraging the novel SwiftUI framework. The research conducts a thorough evaluation of the system’s performance, focusing on user experience, security, and the accuracy of voice and language recognition capabilities.