Bengaluru is slowly becoming an important centre for artificial intelligence in India, and Sarvam AI is yet another name from the city. Founded in 2023, Sarvam AI has focused on building tools that work well with Indian languages from day one. Furthermore, the company has done all its work in India, showing a clear focus on local talent and ideas. The founders say the main goal is to help India become self-reliant in artificial intelligence. Instead of copying global tools, the company wants to solve problems that matter in India. This includes language, voice, and access for everyday users. Here is a clear look at Sarvam AI now overall.
Pratyush Kumar is a graduate of IIT Bombay and holds a Ph.D. from ETH Zurich. He has worked with Microsoft Research and IBM Research. He also teaches at IIT Madras as an adjunct faculty member. Kumar co-founded AI4Bharat, which works on AI tools for Indian languages. He also helped start PadhAI, an online learning platform that aims to make quality education affordable for more people.
Pratyush Kumar and Vivek Raghavan founded Sarvam AI in 2023. Their goal was to build basic AI models made specially for Indian languages.
Many global AI tools struggle with India’s wide mix of languages, scripts, and accents. Sarvam AI was created to fill this gap, as the founders wanted AI to understand difficult Indian scripts, speak in natural voices, and stay affordable. These were hard goals, but they guided the company’s mission from the very beginning.
Also read: Oppo Find X8 Pro available with Rs 19,000 discount: Here’s how to grab this deal
Sarvam AI faced many early challenges. Training AI to read different scripts, understand many dialects, and speak naturally in Indian languages was hard. Each language came with its own rules and sounds.
Cost was another big issue. Most global AI tools are expensive and built for users outside India. Sarvam AI worked to keep its tools affordable and useful for Indian users with limited internet and devices.
Sarvam Vision is the company’s text reading tool. It can read scanned documents and images. It scored 84.3 per cent on the OlmOCR test and 93.28 per cent on the OmniDoc test, which are strong global results.
Bulbul V3 is Sarvam AI’s voice tool. It supports 35 voices across 11 Indian languages. The company plans to cover all 22 official Indian languages in the future.
Sarvam AI is legally called Axonwise Pvt. Ltd. It was chosen by the Indian government to build India’s first sovereign large language model under the IndiaAI Mission. The company received access to around 4,000 NVIDIA H100 GPUs to train this model.
Sarvam AI claims its models perform better than Google Gemini and GPT-4o on Indian language tasks. Sarvam Vision scored higher than DeepSeek and Gemini in reading Indian documents. Its speech model made fewer mistakes, especially with mixed languages like Hinglish.
Unlike many global models that mostly focus on text, Sarvam AI puts strong focus on voice. This helps people like farmers and delivery workers who find speaking easier than typing. Bulbul V3 works well even with low internet and can take voice commands directly. For example, it understands that the word ‘Nau’ means the number nine, not the word ‘no’.