Sarvam Edge (www.sarvam.ai)

Sarvam Edge has been launched to deliver on-device AI capabilities, aiming to enhance accessibility and performance by running sophisticated models directly on users' devices. This initiative emphasizes that the future of AI will not solely depend on the size of models but on how effectively they can be deployed at a national scale, enabling immediate responses without reliance on cloud services. The architecture prioritizes efficiency and privacy, allowing users to process data locally without any server interaction, which is particularly significant for users in areas with limited connectivity. The Sarvam Edge initiative includes advanced speech recognition and synthesis models tailored for the Indian market, supporting multilingual capabilities across ten popular languages. For example, the speech recognition model delivers real-time transcription with a latency of under 300 milliseconds and operates entirely offline, ensuring user privacy. The models are designed to be cost-effective as they eliminate per-query costs associated with cloud services. Furthermore, the translation model supports bidirectional translation across eleven languages using a unified architecture, achieving competitive accuracy with a significantly smaller footprint. This groundbreaking approach not only redefines the accessibility of AI but also opens new possibilities for applications in education, productivity, and assistive technologies, especially in underserved communities.