Achieving human-like, effortless navigation in mobile robots requires a conceptual understanding that extends beyond mere obstacle avoidance. This capability, particularly in Object Navigation (ObjectNav) tasks, demands both a comprehensive spatial awareness of the environment and a deep semantic comprehension of user instructions. Because natural language commands are often unstructured, Large Language Models (LLMs) are uniquely suited to interpret these inputs and extract actionable goals for the robot. However, deploying such high-fidelity perception and spatial reasoning systems on resource-constrained hardware presents significant computational challenges. This thesis presents an end-to-end object navigation pipeline for indoor environment that is optimized for edge deployment, utilizing an LLM framework for high-level behavioral orchestration.