This paper presents a novel architecture for a Hindi voice-based assistant system tailored for e-commerce applications. The architecture integrates key components including speech recognition, machine translation, keyword extraction, intent classification, a dictionary-based question-answering model, and speech synthesis. The system efficiently converts spoken Hindi queries into text, translates across languages, identifies intent, extracts keywords, generates relevant database queries, and provides accurate responses through synthesized speech. This holistic approach enhances user interaction, accessibility, and convenience within the e-commerce domain.