The global rise in the elderly population presents critical challenges in ensuring mental well-being, addressing social isolation, and enabling early detection of cognitive decline. The paper proposes a novel framework leveraging Large Language Models (LLMs) to provide empathetic and intelligent friendship to elderly individuals while monitoring their cognitive and emotional health. Our proposed system integrates a dialogue manager, memory-enhanced interaction, embedded cognitive assessments, sentiment analysis, and a caregiver dashboard. The framework supports personalized engagement via natural conversation, enabling non-invasive, culturally adaptable, and privacy-preserving interactions. It will also include passive interaction using group discussions, incorporating family-related information extracted and updated from social media platforms such as Facebook, Instagram, and WhatsApp etc. The proposed approach holds promise for aiding healthcare providers in the early detection and continuous monitoring of age-related mental conditions such as Mild Cognitive Impairment (MCI), social isolation, and depression.