This thesis will analyze how AI-driven symptom checkers in digital healthcare have changed over time, how they work, and what their ethics may entail. It explores how these tools have evolved to being more basic, rule-based systems to complex machine learning and large language model (LLM) architectures enabling personalized and conversational interactions. The paper focuses on the incorporation of Retrieval-Augmented Generation (RAG) to improve the accuracy of the facts and reliability of the data without jeopardizing the privacy of the user by conducting processing locally and on-the-dollies. With the help of RAG, the system embraces the most recent knowledge bases and, therefore, minimizes hallucinations and enhances the overall credibility of the obtained medical data. Moreover, the study is critical on the ethical, legal, and social issues surrounding the data sharing, misinformation, and user trust in the domain of digital symptom assessment. It is a discussion of possibilities of regulatory frameworks and expectations of the society that can be used in responsible deployment of such technologies. The thesis suggests the evidence-based, privacy-saving, and patient-safe technological innovation based on its final part. This framework will help in responsible AI implementation in medical situations, making them effective and accountable.