Do you know about sarvam ai? And How it’s work?

Leave a Comment / AI / By

Sarvam AI is a Bengaluru-based Indian startup that is building large language models (LLMs) specifically designed for India’s linguistic diversity and unique use cases.

Here’s a breakdown of what makes Sarvam AI significant:

  • Sovereign AI for India: The Government of India has selected Sarvam AI under the IndiaAI Mission to build the country’s first indigenous, sovereign large language model. This means the model will be developed, deployed, and optimized within India, using local infrastructure and talent.
  • Focus on Indian Languages and Voice: Unlike many global AI models that are primarily trained on Western data, Sarvam AI is developing LLMs from the ground up that are proficient in multiple Indian languages. They also have a strong focus on voice-first applications, which is crucial for bridging digital access gaps in a country like India.
  • Founders with Deep Expertise: It was founded in 2023 by Dr. Vivek Raghavan and Dr. Pratyush Kumar, both of whom have significant experience in building digital public infrastructure and pioneering AI applications for Indian languages (including their involvement with AI4Bharat).
  • Full-Stack Generative AI Platform: Sarvam AI offers a full-stack Generative AI platform, including various model variants (Sarvam-Large for advanced reasoning, Sarvam-Small for real-time applications, and Sarvam-Edge for on-device tasks) and products like Sarvam Agents, Sarvam 2B, and Shuka 1.0.
  • Addressing India’s Needs: Their aim is to make generative AI accessible at scale in India, enabling AI to understand and interact in multiple Indian languages, and to unlock secure, scalable AI applications for citizens, enterprises, and the government without sending data beyond borders. They are also developing AI-powered productivity tools for the government, such as analyzing public data and assisting policymakers.

In essence, Sarvam AI is positioned at the forefront of India’s generative AI revolution, working to build an AI ecosystem that is tailored for India’s unique linguistic and operational realities.

How Sarvam AI Works (Under the Hood)

  1. Large Language Models (LLMs) from Scratch: Unlike many AI startups that fine-tune existing global models, Sarvam AI is building its own LLMs from the ground up. This allows them to specifically tailor the models for India’s unique linguistic and cultural nuances.
    • Data Training: They train their models on massive datasets that include a significant portion of Indian language data (unlike global models where Indian languages might be underrepresented). For instance, their Sarvam 2B model was trained on a proprietary dataset of 4 trillion tokens, with about 40% of those tokens being Indian language tokens. This efficient tokenization helps the models understand and generate Indian languages more effectively.
    • Multilingual and Voice-First: Their models are designed to be fluent in multiple Indian languages, including the ability to understand and generate “code-mixed” conversations (a blend of English and an Indian language, common in India). They also prioritize voice-based interactions, recognizing the importance of voice for digital inclusion in India.
    • Model Variants: Sarvam AI develops different model variants to cater to various needs:
      • Sarvam-Large: For advanced reasoning and complex content generation.
      • Sarvam-Small: Optimized for real-time, interactive applications requiring speed and responsiveness.
      • Sarvam-Edge: For compact, on-device processing, enabling AI capabilities on mobile and IoT devices.
  2. Full-Stack Platform: Sarvam AI provides a comprehensive platform that covers the entire Generative AI development and deployment lifecycle. This includes:
    • Foundational Models: The core LLMs they build.
    • APIs (Application Programming Interfaces): These are the primary way developers interact with Sarvam AI’s models. They provide endpoints for various AI tasks.
    • Tools and Services: They offer various tools and services to help enterprises and developers integrate and deploy their AI solutions.
  3. Infrastructure and Sovereignty: A key aspect of Sarvam AI’s work is its commitment to data sovereignty. Their models are built, deployed, and optimized within India, often leveraging local compute infrastructure (like Yotta’s Shakti Cloud, which uses NVIDIA AI Enterprise software). This ensures data remains within India’s borders, crucial for government and critical enterprise applications.

How to Use Sarvam AI

Sarvam AI primarily provides its capabilities through APIs (Application Programming Interfaces) and, in some cases, pre-built “agents” for specific use cases. This means it’s generally used by developers and businesses to integrate AI functionalities into their own applications, rather than by end-users directly interacting with a consumer-facing product like ChatGPT.

Here’s a general overview of how it’s used:

  1. Obtain API Key:
    • You would typically sign up on the Sarvam AI dashboard to create an account.
    • Upon signing up, an API key is generated for your account. This key is essential for authenticating your requests to their services. It’s crucial to keep this key secure and never expose it publicly.
  2. Install SDK (Software Development Kit):
    • Sarvam AI provides SDKs (Software Development Kits) in popular programming languages like Python and JavaScript.
    • You would install the relevant SDK in your development environment (e.g., pip install sarvamai for Python).
  3. Interact via APIs:
    • Once the SDK is installed and your API key is configured (usually as an environment variable), you can make calls to Sarvam AI’s various API endpoints.
    • Common API Use Cases:
      • Text Translation: Translate text from one language to another (supporting multiple Indian languages and English).
      • Speech-to-Text (STT): Convert spoken language into written text (with support for various Indian languages).
      • Speech-to-Text Translate: Combine speech recognition and translation, converting spoken language directly into translated text.
      • Text-to-Speech (TTS): Convert written text into natural-sounding spoken words (with multiple voice options and Indian language support).
      • Transliteration: Convert text between different scripts while maintaining the same language.
      • Language Identification: Automatically detect the language of input text.
      • Sarvam Agents: These are pre-built, voice-enabled, multilingual, and action-oriented AI agents designed for specific business applications (e.g., customer service via telephone, WhatsApp, or in-app). Businesses can deploy and customize these agents.
  4. Integrate into Your Application:
    • The results from Sarvam AI’s APIs (e.g., transcribed text, translated speech, generated text) are then used within your own application.
    • Examples:
      • A customer service application could use Sarvam AI’s Speech-to-Text and Text-to-Speech APIs to enable voice-based interactions in Hindi, Tamil, or other Indian languages.
      • A content creation tool might use their text translation capabilities for localized content.
      • A government service portal could leverage Sarvam Agents for multilingual voice support for citizens.

In summary, Sarvam AI works by building and deploying specialized AI models for India’s linguistic landscape, and it’s used by developers and enterprises to integrate these capabilities into their products and services via easy-to-use APIs.

Leave a Comment

Your email address will not be published. Required fields are marked *