India’s AI Ambitions Take Shape with Shivaay, But Questions Linger

India’s artificial intelligence (AI) landscape is witnessing a pivotal moment with the emergence of Shivaay, a 4-billion-parameter foundational AI model developed by 20-year-old entrepreneur Rudransh Agnihotri and his team at FuturixAI and Quantum Works. Positioned as a breakthrough in India’s AI journey, the model has sparked both excitement and controversy, raising questions about originality, scalability, and the nation’s broader AI ambitions.

Agnihotri describes Shivaay as a step toward technological self-reliance, aiming to build an AI model that understands India’s linguistic and cultural nuances better than foreign counterparts. However, the project faced significant challenges, with the team—primarily college undergraduates—operating without venture capital funding. Instead, they relied on support from initiatives like NVIDIA Inception and AWS Activate, managing to train Shivaay on just eight NVIDIA A100 GPUs. By leveraging synthetic data annotation techniques and drawing insights from Meta’s Lima paper, they optimized the training process with a focus on high-quality, stylized prompts.

Despite limited resources, Shivaay has delivered promising results, achieving 91.04% on the ARC-Challenge benchmark for reasoning tasks and 87.41% on GSM8K for arithmetic reasoning. However, its release has not been without scrutiny. Following the publication of benchmark results, critics speculated that Shivaay might be a fine-tuned derivative of existing open-source architectures like Llama or Mistral. The skepticism intensified after a system ROM leak revealed references to other models in Shivaay’s internal code.

Addressing these concerns, Agnihotri maintains that while his team incorporated knowledge from multiple models, they built Shivaay’s architecture from the ground up. “This wasn’t about copying; it was about learning from what works while tailoring it for Indian contexts,” he stated, emphasizing transparency by publishing evaluation scripts and benchmarks.

A key differentiator for Shivaay is its ability to process multiple Indic languages, including Hindi, Tamil, and Telugu. This aligns with India’s larger AI vision under the IndiaAI Mission, which aims to develop indigenous foundational models by 2026. The government has allocated ₹10,372 crore for AI infrastructure and plans to deploy 18,000 GPUs nationwide. However, Agnihotri argues that high-performance computing remains inaccessible to most startups, citing the high costs of GPU access as a major barrier.

Taking cues from China’s DeepSeek, which built a high-performing AI model on limited resources, Agnihotri’s team is now exploring advanced optimization techniques, such as FP8 quantization and mixed-precision training, to improve Shivaay’s efficiency. Looking ahead, the team envisions Shivaay evolving into an agentic AI capable of structured reasoning and task-specific applications. Discussions are already underway with IITs to explore AI-driven solutions in agriculture and healthcare, such as early disease detection and precision farming.

While India still faces hurdles in AI research—particularly a shortage of engineers specializing in foundational models—Agnihotri remains optimistic. He believes that with the right industry-academia collaborations and government support, India can carve out a niche in AI innovation by focusing on domain-specific, compute-efficient models rather than competing with the scale of GPT-4. As the AI race intensifies, Shivaay represents both a bold experiment and a test of India’s ability to shape its own digital future.

- Advertisement -

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Latest Articles

error: Content is protected !!

Sign Up for CXO Digital Pulse Newsletters

Sign Up for CXO Digital Pulse Newsletters to Download the Research Report

Sign Up for CXO Digital Pulse Newsletters to Download the Coffee Table Book

Sign Up for CXO Digital Pulse Newsletters to Download the Vision 2023 Research Report

Download 8 Key Insights for Manufacturing for 2023 Report

Sign Up for CISO Handbook 2023

Download India’s Cybersecurity Outlook 2023 Report

Unlock Exclusive Insights: Access the article

Download CIO VISION 2024 Report

Share your details to download the report

Share your details to download the CISO Handbook 2024

Fill your details to Watch