Gnani.ai Focuses on Partnership Models and Developing New SLMs Going forward, Gnani.ai is also looking to provide its voice SLM solutions to sectors, particularly automotive and healthcare and is looking to put more of its voice SLMs on edge computing

After six years of exploring voice-based generative artificial intelligence (GenAI), Gnani.ai is now set to focus on a new sub-set of AI- Small Language Model (SLM). Additionally, while the US and India continue to remain a key geographical priority, it is planning to build partnership models for countries outside the said markets.

Speaking to BIZ Experiences India, Ganesh Gopalan, Co-Founder and CEO, Gnani.ai said, "This is a work in progress with a lot of new things yet to come. We will focus on SLMs along with the voice AI."

The Bengaluru-based start-up has been offering its comprehensive suite of conversational AI products to automate customer interactions across various industries. Gnani.ai launched India's 1st voice-first SLM for vernacular languages early this year and raised USD four million in a Series A funding round from Info Edge Ventures.

Developing solutions from a new technology requires tinkering and that's what Gnani.ai is doing. It began using open-source products such as OpenAI, Llama, Mistral and other large LLMs for its customers.

"Slowly, we realised that there are issues pertaining to these larger models which are built for very generic complications. You ask them anything and they can answer. But that comes with its own drawbacks, too many hallucinations, issues with latency, and high compute costs or deployment costs. So that's when we decided to build our own SLMs. We are not here to compete with these large LLMs, we are here to solve problems for end customers which many of these LMS are not solving," Ganesh Gopalan, Co-Founder and CEO, Gnani.ai told BIZ Experiences India.

Trained on millions of audio hours of proprietary audio datasets and billions of Indic language conversations, its SLM captures diverse dialects, accents, and linguistics across the country. Gnani.ai's SLM data size is relatively smaller compared to LLMs such as GPT-4 and Llama-3. "Our latencies are less than 200 milliseconds. So, it's like having an actual conversation," he added.

SLMs are trained on smaller, more focused datasets than LLMs. They use domain-specific and proprietary data to give efficient and accurate responses. It is developing voice SLM catering to specific industries and at present, its SLM is deployed in the banking, financial services, and insurance space. According to reports, the company has built a series of five models designed for the BFSI sector. For catering to banking players, Gnani.ai fine-tunes its SLM with bank-specific vocabulary to engage with customers, "That really shoots up the accuracy. With low latency, high accuracy, and low hallucinations, we are happy the way it has come across," Gopalan said.

Going forward, Gnani.ai is looking to provide its voice SLM solutions to sectors, particularly automotive and healthcare and is looking to put more of its voice SLMs on edge computing.