Red Hat and Hugging Face Push India to Build AI, Not Just Use It

In Bengaluru, Red Hat and Hugging Face brought together more than 170 engineers, students, and researchers for a working session on the machinery behind modern AI. The message, set early by Red Hat's Sudhir Dharanendraiah, was simple: India is a huge user of AI, but it should also become a serious creator of AI infrastructure. That means moving beyond simply calling models or tweaking them, and instead shaping the tools that many other people depend on. For an audience of early-career professionals, this signals a new class of high-level opportunities in the tech industry.

One talk focused on profiling, which means measuring where a program spends its time before trying to make it faster. The speaker, Aritra Roy Gosthipaty, stressed a practical rule: you can't improve what you don't measure. He showed that a small AI job may seem slow even on good hardware, not because the AI itself is slow, but because the computer overhead of starting the job takes up most of the time. The lesson helps developers avoid guessing and instead find real bottlenecks. This applies beyond engineers — any tool that responds slowly may be suffering from wasted time outside the main task.

Another talk tackled serving AI to many users at once. It's easy to make a demo for one person; it's much harder to keep an AI fast when thousands of people ask questions at the same time. Technologies like SGLang and improved kernels help manage these heavy loads. This is the behind-the-scenes layer that decides whether a chatbot feels instant or sluggish on a busy day. Better systems here mean lower costs for companies and a smoother experience for everyone.

The event highlighted the growing idea that India's AI future is not just about using AI made elsewhere. With local talent working on profiling tools, serving engines, and training systems, the entire global AI ecosystem becomes healthier and more diverse. For regular users, this could lead to AI that understands local languages, fits local needs, and costs less. The honest limitation is that building core infrastructure is hard, slow work requiring deep expertise — but gatherings like this are exactly how that expertise begins.