-1770588711440.webp)
Sarvam AI is an innovative Bengaluru-based start-up, which was founded in 2023 and is focused on developing the full-stack solutions of generative AI that will be adapted to the diversity of the Indian language.
Having a focus on voice-enabled models with support of more than 10 Indian languages, it covers cultural nuances and code-mixing to address sovereign AI innovations.
When was Sarvam AI Founded?
In August 2023, Sarvam AI was established by Indian digital public infrastructure expert Dr. Vivek Raghavan and Dr. Pratyush Kumar, the founder of open-source Indian languages AI4Bharat, an IIT Madras lab.
Their joint vision was formed by the understanding of the gap in AI models that are appropriate to comprehend the reality of the 1.4 billion population of India with its 22 official languages and a lot of dialects.
This two-person start-up funded the company with a vision of developing an AI that was accessible and India centric at the core.
What is the Legal Name of Sarvam AI?
As an AI-based company under the legal name of Axonwise Private Limited, Sarvam AI intended to democratize AI by building core models that were trained on trillions of India-specific tokens.
Early objectives revolved around addressing weaknesses of Western-centric LLMs, including inaccurate treatment of idiomatic expressions, transliterations and code-switching of languages that are prevalent in Indian communication.
The goal of the company was efficiency in edge devices and complete data sovereignty in India.
Where are the Headquarters of Sarvam AI?
Based in Bengaluru, Karnataka ( White Field Hosakote Main Road ) Sarvam AI is in the heart of India and its growing tech ecosystem. Bengaluru is located near the highest research centers such as IISc and a huge amount of AI talent, which renders it ideal in terms of rapid-iteration and scaling.
The option indicates strategic availability to compute infrastructure and alliances in the Silicon Valley of India.
What are the Key Focus Areas of Sarvam AI?
Sarvam AI is, most fundamentally, a company that focuses on multilingual large language models (LLM) that are trained on varying datasets, such as Hindi, Tamil, Telugu, Malayalam, Punjabi, Odia, Gujarati, Marathi, Kannada, and Bengali, among others.
The focus is put on voice-first interfaces, mobile/edge deployment low-latency models, and Indian data localization legislative-compliant. Some of the innovations involve addressing regional accents, colloquial speech, and real time translation.
Mission Pillars
The mission of Sarvam is based on four pillars:
(1) Voice-based, multilingual AI to have natural interactions
(2) Compact and customizable models to serve enterprises
(3) Action-oriented AI agents to be practical in use
(4) Sovereign AI infrastructure to preserve the data privacy and ethical development. These pillars are used to open-source work and partnerships to create an Indian AI stack.
Sarvam AI’s Flagship Open-Source Models
Sarvam has published various pioneering open-source models, the first being Sarvam 2B, the first 2-billion-parameter Indic LLM ever to be trained entirely on 4 trillion tokens using domestic compute hardware.
Sarvam-1 is the best code-mixing and efficiency code-switching framework in 10 Indian languages, which is supported by NVIDIA NeMo.
Shuka v1/1.0, the first free-software Indic speech LLM, takes audio as input to text output, and is the first to be benchmarked better than state-of-the-art, with Hindi and other languages supported through Llama-3 architecture.
Voice and Speech Tools
Sarvam's voice suite transforms interactions for Indian users:
| Product | Description | Languages Supported |
| Bulbul 1.0 | High-fidelity text-to-speech with 6 expressive voice options for natural prosody | 10 Indian languages |
| Saaras 1.0 | Robust speech-to-text with auto-language detection, diarization, and translation | 10 Indian languages |
| Saarika | Specialized speech-to-text optimized for Indic accents and noisy environments | Major Indian languages |
| Mayura 1.0 | Advanced translation API handling formal, colloquial, and code-mixed Indic text | 10+ Indian languages |
Data Source: Sarvam
Key Collaborations
The strategic alliances only enhance the reach of Sarvam: Microsoft incorporates the Sarvam Indic Voice LLM on Azure to provide scalability to the enterprise. Training is powered by Yotta Cloud infrastructure based on Shakti.
The financial services, legal firms, consumer goods, tech giants, media and telecom are all covered in partnerships and pilots in voice banking and customer support have been tried. The support of the government through IndiaAI Mission highlights its contribution to AI sovereignty in the country.
Sarvam helps in the open-source maintenance of data, benchmarks against new SOTA in Indic tasks and serves millions of real-world deployments. Long-term objectives are 100M+ users through edge AI and worldwide Indic growth.
Check Out: Which Palace Is Called the Palace of Winds?
-1770611089930.jpg)
Comments
All Comments (0)
Join the conversation