Healthcare accessibility in regions with limited internet connectivity remains a critical challenge, as traditional telemedicine solutions are heavily reliant on online infrastructure. This paper presents Viksith, an innovative offline multimodal medical assistant designed to provide medical guidance through speech, text, and image inputs without requiring internet access. Viksith employs resource-efficient algorithms, including locally optimized machine learning models and rule-based decisionmaking, to deliver accurate and timely medical support on low-power devices. The architecture ensures compatibility with resource-constrained environments, making it ideal for rural and underserved areas. Comprehensive testing and pilot deployments demonstrate Viksith’s high performance in recognizing symptoms, analyzing visual inputs, and generating actionable medical insights. This paper provides a detailed exploration of Viksith’s design, implementation, and evaluation, positioning it as a scalable solution for bridging healthcare gaps in offline scenarios.