Meet Pratyush Kumar: The CEO of Sarvam AI, India’s homegrown startup giving Google and ChatGPT a run for their money

India is consistently making forward strides in its AI game, and while one would expect a boom in technology from bigger, well-known giant names, the ones actually making a mark are startups.One among these is Sarvam AI from Bengaluru, proving the country can hold its own against giants from the US and China.

What is Sarvam AI?

Sarvam AI is a Bengaluru startup founded in 2023 that has pioneered foundational AI models built entirely in India. Their key trait is working on important technologies like optical character recognition (OCR) and text-to-speech for Indian languages, some fields where major global companies have fallen short.

Their vision for "sovereign AI" is about building self-dependent, India-focused systems that match world-class performance while addressing local priorities.

Meet Pratyush Kumar- CEO Sarvam AI

Pratyush Kumar co-founded Sarvam AI along with Vivek Raghavan, driving its vision for self-dependent AI. He previously launched AI4 Bharat for Indian language AI apps and PadhAI for affordable online learning. He has a Ph.D. from ETH Zurich, B.Tech from IIT Bombay, and experience at Microsoft Research, IBM Research, and IIT Madrasm where he's adjunct faculty, and shares progress actively on X.|

Why Sarvam's vision outshines others

Sarvam Vision, their OCR tool, leads the pack on the olmOCR-Bench with an impressive 84.3% accuracy score, outpacing models like Gemini 3 Pro, DeepSeek OCR v2, and ChatGPT. It also scored 93.28% overall on OmniDocBench v1.5, standing out in tricky areas like complex layouts, technical tables, and math formulas where traditional OCR often falters. Users love its trustable performance on everyday real-world documents.

‘Bulbul’ Speaks Indian

Bulbul V3, another tool from Sarvam AI, provides natural-sounding text-to-speech across 11 Indian languages with 35 distinct voices, and they're aiming to expand to 22 languages total. It is capable of competing with the quality of ElevenLabs but at much more affordable prices, especially for Indic applications. KissanAI founder Pratik Desai shared on X, “We use Bulbul as our go-to TTS model for our Indic use cases, and they have just gotten better with each release.

Meanwhile, ElevenLabs cost never made sense for Indic or any other languages.”

Sarvam AI has got appreciation from tech experts

Skeptics like tech commentator Deedy Das, once doubtful about Indic focus, now praise them. On X, Das wrote: “They have the best text-to-speech, speech-to-text, and OCR models for Indic languages, and that’s actually really valuable. The pricing is very reasonable.”