Generative AI LLM agent preserve user privacy and safety

Generative AI has fundamentally revolutionized the way humans interact with computers, enabling intelligent and personalised assistance that boosts productivity, across all facets of human lives. However, organizations worldwide have been imposing general ban on popular AI data tools like ChatGPT and Bard due to serious questions regarding data privacy and security. This work thesis suggests new approach by creating an open-source privacy-preserving "Generative AI Agent" desktop framework to use and develop further with no data sharing of any kind. The research aims to develop an AI safe-language using Large Language Models (LLM) and Retrieval Augmented Generation (RAG), providing personalized assistance while protecting proprietary information. Further, the work suggests a simple tool to spot privacy breaches in HCI along with ways of addressing them. The main contributions are customizable generative AI system to be run on-premises and detection of privacy breaches. The focused on improve the privacy and security of generative models, with improved practices in local secure deployment as well. The experiment results shows that the designed agent when used with high-level personal data, show high private services and assessing the safety bias on privacy issues.