Privacy-Preserving Real-Time Vietnamese-English Translation on iOS using Edge AI

This research addresses the growing need for privacy-preserving and accessible language translation by developing a fully offline Neural Machine Translation (NMT) system for Vietnamese-English translation on iOS devices. Given increasing concerns about data privacy and unreliable network connectivity, on-device translation offers critical advantages. This project confronts challenges in deploying complex NMT models on resource-limited mobile devices, prioritizing efficiency, accuracy, and a seamless user experience. Leveraging advances such as MobileBERT and, specifically, the lightweight TinyLlama 1.1B Chat v1.0 in GGUF format, a quantized Transformer-based model is implemented and optimized. The application is realized as a real-time iOS prototype, tightly integrating modern iOS frameworks and privacy-by-design principles. Comprehensive documentation covers model selection, technical architecture, challenges, and final implementation, including functional Swift code for deployment.