To address the gap in artificial intelligence (AI) compute accessibility in India, IIT-Madras and its tech arm, IITM Pravartak Technologies Foundation, have collaborated with California-based deep-tech startup Ziroh Labs to launch a Centre of AI Research.
Indigenously developed, Kompact AI enables the training and deployment of foundational models using standard CPUs, making high-performance AI more accessible, particularly in resource-constrained environments. The platform has already been used to optimise models like DeepSeek, Qwen, and Llama for CPU-based inference, with benchmarking support from IIT-Madras researchers.
Further, the new Centre will focus on building compact, task-specific AI models suitable for deployment in rural and underserved regions, where high-end computing infrastructure is often limited. Notably, Kompact AI can function without continuous internet connectivity and is being positioned as a transformative solution for sectors such as agriculture, education, and healthcare, where low-cost, on-device AI can have significant impact.
Furthermore, Ziroh Labs claims that Kompact AI delivers up to three times the performance of existing CPU-based AI frameworks without sacrificing accuracy, overcoming the limitations typically associated with model quantisation and distillation efforts.
