By: Tom White

Creating AI voice agents is a growing industry, with many companies and individual users paying for premium, artificially generated voices for applications ranging from video games to audiobooks to customer service bots. In this field, founders Sudarshan Kamath and Akshat Mandloi have carved out an unusual niche for their platform Smallest.ai—they are the only provider offering free voice agents.

Making Voice AI Accessible at Scale

In a bold move to make AI more accessible, Smallest.ai provides hundreds of hours of free voice calling to everyone in America. This initiative aims to democratize the technology and make high-quality AI voice agents available to the general public, driving the idea of “frugal innovation” to a broader audience. By offering this free access, the company hopes to encourage more individuals and businesses to experiment with AI voice technology, accelerating its adoption and development across various sectors.

Where Smallest.ai Came From

Smallest.ai is an AI research company specializing in small, multimodal language models. Founders Sudarshan and Akshat became convinced that smaller models were the future after years spent optimizing larger models to work with limited hardware while developing perception algorithms for Level 3 self-driving vehicles in several countries.

Both founders are IIT Guwahati (IITG) graduates, one of the competitive technical institutions. Before their work on self-driving vehicles, Sudarshan was part of the core team at Toppr.com, an online learning platform acquired for $150 million. Meanwhile, Akshat was among the first employees at Tork Motors, India’s first electric bike company. Now, they have applied their expertise in self-driving vehicle data science and adapting large language models to accelerate machine learning.

Photo Courtesy: Smallest.ai / Akshat Mandloi & Sudarshan Kamath

Can Small Models Can Outperform Larger Models?

Having seen firsthand the limitations of large language models, Sudarshan and Akshat believe that the current large neural networks are over-parametrized and that to achieve artificial general intelligence (AGI), what is needed are much smaller AI models that focus on core algorithmic innovations. That is the thinking behind Smallest.ai.

Customers and Investors are both surprised by the swift success the company has seen, especially with smaller models.In under six months and with a budget of less than $100,000, Smallest.ai outperformed several larger players in third-party open-source Text-to-Speech (TTS) benchmarks. One of these companies is valued in the billions.

Smallest.ai states that it has developed a voice cloning pipeline designed to replicate the distinct speech patterns of individuals with a high level of accuracy. This allows Smallest.ai to provide custom voice agents for calling operations ranging from outbound sales, lead qualification, and debt collection to customer support. Starting with an initial $100,000, the company has since raised significant funding from various investors, including those with early stakes in major tech companies.

It is cost-effective AI, creating premium-quality voice-to-voice AI models at less than a third of the cost of Smallest.ai’s competitors. Smallest.ai is also technologically independent, relying on its own models without using large models developed by OpenAI, Anthropic, or other large AI labs. This allows Smallest.ai to iterate and adapt more quickly because they are training much smaller models and doing so in-house. 

Today, Smallest.ai provides real-time voice agents, text to speech services to over 10,000 small and medium-sized businesses worldwide. Their voice agent platform, Atoms, also offers one of the lowest-latency, lowest-cost AI-driven voice services currently on the market.

Looking Ahead

Sudarshan and Akshat are dedicated to “first principles thinking,” challenging developers to rethink fundamental assumptions about AI scalability and accessibility. By providing free voice agents outperforming larger models, they seek to get more people thinking about “frugal innovation” for AI. Smaller models allow for more efficient capital allocation in AI, which, Sudarshan and Akshat note, lowers the barrier to entry “for anyone who wants to get their hands dirty with training AI models.”

As for the future of Smallest.ai, the founders hope to see one billion humans speaking to Smallest.ai models daily. “To enable this,” Sudarshan and Akshat say, “we need to start thinking of models as a finite-state machine, not a single input/output module. Something that continuously runs, updates and improves … a system that constantly learns and understands from its interactions, making it more useful as it interacts with humans.”

Is the future of AI to be found in smaller, swifter-to-adapt models? It may be an intriguing possibility.

Disclaimer: This article is for informational purposes only and does not constitute financial, investment, or technical advice. The views expressed are those of the founders and do not necessarily reflect the opinions of any affiliated organizations. While efforts have been made to ensure accuracy, AI technology and market dynamics evolve rapidly. Readers should conduct independent research before making business, financial, or technological decisions related to AI voice agents or similar innovations.

Published by Joseph T.