India is a country known for its incredible linguistic diversity. With 22 officially recognized languages and over 19,000 dialects, the country presents a unique challenge for businesses and service providers aiming to reach its vast population. In such a linguistically fragmented market, offering a text-only AI chatbot that functions efficiently in just a few languages could seem inadequate. The question arises: does it make sense to rely solely on text-based communication in such a context, or should more inclusive and versatile solutions be explored?
Sarvam AI: Addressing the Language Barrier
Indian AI startup Sarvam has been grappling with this very issue. On a mission to make AI more accessible to India’s multilingual population, the Bengaluru-based company has developed a suite of offerings that includes a voice-enabled AI bot supporting over 10 Indian languages. This strategic move acknowledges a crucial insight: many Indians would rather speak to an AI in their native language than engage in text-based communication.
On Tuesday, Sarvam unveiled this series of offerings, betting on the power of voice interaction to break down the barriers imposed by linguistic diversity. This voice-enabled AI bot is designed to function seamlessly across various platforms, including WhatsApp, apps, and traditional voice calls, making it a versatile tool for industries that rely heavily on customer support.
Why Voice-Enabled AI?
The preference for voice interaction over text is not unique to India, but it is particularly pronounced in a country where typing in local languages remains a significant challenge. While English and Hindi are widely used, many people are more comfortable speaking in their regional languages. This is where Sarvam’s voice-enabled AI bot steps in, providing a solution that aligns with the cultural and linguistic realities of the Indian market.
“People prefer to speak in their own language. It’s extremely challenging to type in Indian languages today,” said Vivek Raghavan, co-founder of Sarvam AI, in a recent interview with TechCrunch. This insight drives the startup’s approach to AI development, prioritizing voice interaction as a more natural and efficient means of communication.
A Multilingual Solution for Businesses
Sarvam’s AI voice agents are tailored for a range of industries, with a particular focus on those that require robust customer support systems. One notable example is Sri Mandir, a startup offering religious content, which has integrated Sarvam’s AI agent to handle transactions. To date, the AI has successfully processed over 270,000 transactions, demonstrating the practical utility of voice-enabled AI in the Indian context.
The pricing model for Sarvam’s AI agents is another aspect that underscores the company’s understanding of the Indian market. By offering AI services at just ₹1 (approximately 1 cent) per minute of usage, Sarvam makes its technology accessible to a broad range of businesses, from small startups to large enterprises.
The Technology Behind Sarvam AI
At the core of Sarvam’s offerings is its small language model, Sarvam 2B. This model is built on a data set of 4 trillion tokens, entirely generated from synthetic data. The use of synthetic data, while controversial, was a deliberate choice due to the scarcity of Indian language content available on the open web.
Synthetic data, which is essentially data generated by large language models to mimic real-world data, is often met with caution by AI experts. The concern lies in the potential for “hallucinations,” where the AI generates information that is not accurate or based on real data. Training AI models on such data can exacerbate these inaccuracies, leading to potential reliability issues.
However, Sarvam has implemented measures to mitigate these risks. The startup has developed models specifically designed to clean and enhance the synthetic datasets, ensuring that the AI’s outputs are as accurate and reliable as possible. Despite the challenges associated with synthetic data, Sarvam 2B is positioned as a cost-effective alternative to larger language models, with the company claiming it will be ten times cheaper than comparable industry offerings.
Open-Sourcing for Innovation
In a move that reflects its commitment to fostering innovation, Sarvam has opted to open-source its language model. By making Sarvam 2B available to the broader community, the startup hopes to encourage further development and adaptation of its technology. This open-source approach not only democratizes access to advanced AI tools but also accelerates the pace of innovation by allowing developers to build upon Sarvam’s foundational work.
“While the large language foundational models are very exciting, you can achieve an experience that is superior, more specific, lower-cost, and with reduced latency using small language models,” Raghavan explained. For use cases that require frequent interactions, such as customer support, Sarvam believes smaller models are more suitable due to their efficiency and cost-effectiveness.
Expanding the AI Toolkit
In addition to its voice-enabled AI bot and small language model, Sarvam is also launching an audio-language model named Shuka. Built on the Saaras v1 audio decoder and Meta’s Llama3-8B Instruct, Shuka is designed to enhance voice interfaces by providing advanced translation, text-to-speech (TTS), and other audio processing capabilities. Like Sarvam 2B, Shuka will be open-sourced, allowing developers to leverage its capabilities for a wide range of applications.
Another innovative product in Sarvam’s portfolio is “A1,” a generative AI workbench tailored for the legal industry. This tool is designed to assist lawyers with various tasks, including looking up regulations, drafting documents, redacting sensitive information, and extracting relevant data. By automating these tasks, A1 aims to increase efficiency and accuracy in the legal profession, freeing up time for lawyers to focus on more complex aspects of their work.
Aligning with India’s AI Ambitions
Sarvam’s work is aligned with broader national efforts to develop a sovereign AI infrastructure in India. Governments around the world are increasingly investing in AI systems that are developed and controlled at the national level, with the goal of safeguarding data privacy, stimulating economic growth, and ensuring that AI development is culturally relevant. India is no exception, with initiatives like the IndiaAI program aiming to build a robust AI ecosystem tailored to the country’s unique needs.
One of the key initiatives under the IndiaAI program is the IndiaAI Compute Capacity project, which aims to establish a supercomputer powered by at least 10,000 GPUs. This initiative reflects India’s ambition to become a global leader in AI, with a particular focus on language-specific models that cater to the country’s diverse linguistic landscape.
Sarvam is well-positioned to contribute to these national efforts. With its focus on developing AI models that are both cost-effective and linguistically inclusive, the startup is already playing a role in advancing India’s AI capabilities. “If the opportunity arises, we will work with the government,” Raghavan said, signaling Sarvam’s readiness to collaborate on national AI projects.
The Future of AI in a Multilingual World
Sarvam’s journey underscores the importance of developing AI technologies that are inclusive and adaptable to diverse linguistic contexts. In a country as linguistically rich as India, offering a one-size-fits-all solution is unlikely to meet the needs of the entire population. By focusing on voice-enabled AI, small language models, and industry-specific tools, Sarvam is addressing a critical gap in the market.
As AI continues to evolve, the need for multilingual and culturally sensitive solutions will only grow. Sarvam’s work serves as a model for how startups can innovate in this space, offering products that are not only technologically advanced but also aligned with the cultural and linguistic realities of their target markets.
The success of Sarvam’s offerings will likely depend on their ability to scale and adapt to the diverse needs of Indian businesses and consumers. However, with a clear understanding of the challenges posed by linguistic diversity and a commitment to making AI accessible to all, Sarvam is well on its way to becoming a key player in India’s AI landscape.
In conclusion, the question of whether it makes sense to offer a text-only AI chatbot in a market as diverse as India has been answered by Sarvam’s innovative approach. By betting on voice-enabled AI and small language models, the startup is not only meeting the needs of its customers but also paving the way for a more inclusive and accessible AI future in India. As Sarvam continues to develop and refine its offerings, it will undoubtedly play a significant role in shaping the future of AI in multilingual markets around the world