Indian startups, alongside the IndiaAI Mission, are focused on creating population-scale impact using AI and exploring use cases that cater specifically to the needs of India. 

On a similar bid, Bhavish Aggarwal, founder of Krutrim, launched Kruti—the country’s first agentic AI assistant capable of executing tasks such as booking cabs, ordering food, paying bills, generating images, and carrying out in-depth research. 

The tool supports both text and voice input, includes a read-aloud feature, features 11 Indic languages, and is built on top of Krutrim V2 LLM, which consists of 12 billion parameters. 

On the sidelines of Kruti’s launch event, AIM had the opportunity to catch up with the three leaders behind the project at the Ola campus. Chandra Khatri, founding head of AI; Navendu A, SVP and head of business; and Sunit Singh, SVP of product at Krutrim, spoke in depth about the team’s future plans and the upcoming tools in the pipeline.

To start off, Navendu clarified a common misconception, explaining that Krutrim V2 is built on Mistral as an architecture, not as a fine-tuned version of it. “Every company has some or the other base architecture,” he said, adding that GPT, Llama, and Mistral are built on top of Google’s Transformer, and that the Krutrim team aims to be transparent with its technical decisions.

“We have made it open source. You can go and see what we have trained,” the team added, emphasising that instead of the model, they only took the architecture and then trained it from scratch using their own multilingual datasets.

Full-Stack Vision

During the launch, Khatri claimed that while most AI assistants in the market focus on narrow tasks or English-speaking users, Kruti is “built for Bharat”. 

He further revealed that Kruti isn’t going to remain a consumer-facing tool. In the coming months, the team plans to launch its SDK and API for developers and enterprises to build their own ‘Krutis’. “If I run an IT company or an e-commerce platform, I can customise my own Kruti using the same base platform,” he said.

Kruti supports over 11 Indian languages and can process India-specific visual inputs like Aadhaar, PAN cards, utility bills, and more. The idea is to create an assistant that can cater not only to users familiar with English and digital UIs but also those who interact with tech in their native language—and often for the first time.

Currently, the team claims to have 90% accuracy for agentic use cases.

The team is also building voice, image, and several other models. For example, Dhwani and Chitrarth, built from scratch, were designed as part of the Krutrim model family, embedding the same values and capabilities. 

“Every time a new model comes, you basically have a corresponding image or voice model,” Navendu said.

The team’s training journey started with Krutrim 1—a demonstration of their ability to build a foundation model in India, even though the infrastructure used then was foreign. With Krutrim 2, they trained a larger model using their own cloud. 

“Was it at the stage of DeepSeek or OpenAI? No. Because the size of our infrastructure was smaller,” Navendu admitted. “We were looking for very specific use cases.”

While the original model supports 11 languages, Krutrim plans to scale to 22 soon. “Twenty-three will go in that direction,” Khatria confirmed. The company believes AI built for India must be inclusive, covering the country’s linguistic, cultural, and digital diversity.

It Takes a Village

Speaking on the current Sarvam controversy around their first model release under the IndiaAI Mission, Navendu said that everyone should be allowed to grow and that this is where the ecosystem can help. “Everyone should have their own time as everyone is doing this for the first time,” he said, while adding that Krutrim has an edge as the team started in 2023 and built smaller models. 

“They [Sarvam] are starting now, and will definitely have their own journeys,” Navendu said, adding that all the startups are working together to build AI use cases for India. “There are 800 million smartphone users in India, and only 200 million are in English. The rest are vernacular users. It’s not about English. Language is just one parameter. It’s also about preferences, thought processes, and cultural nuances. We are building for a complete India.”

The team strongly believes in collaboration, especially with public and academic initiatives like Bhashini and AI4Bharat. “This is a very big problem. Not one company or government can solve it. We will collaborate with everyone,” Navendu further said. “We are talking to IITs. We work with them closely. No one company can solve this, at least at the Indian scale.”

Indian Models for Indian Minds

Krutrim’s commitment to the Indian context is not just technical—it’s ideological. The team emphasised that Indian models must reflect Indian thinking, not just translated English. “Are we talking about Indian thought processes or US thought processes? Models have those nuances also. We are taking care of all of them,” Navendu said.

The team mentioned that much of Krutrim’s work is built with Indian developers in mind. While some critics noted that their app’s UX looks similar to ChatGPT or Perplexity, they see it differently. 

“All mobile phones look like Apple. It’s about familiarity. People are used to it. So why reinvent the wheel unnecessarily?” Singh added. “We have tried our best to bring in lots of colour, to make it feel magical. Most AI apps are too functional. We’ve added our own personality.”

Nevertheless, they understand the challenge. “We’re always thinking about how more people can adopt what we are building and how we can build more use cases to make it super relevant to everyday lives.”

For the Krutrim team, it’s not about global rankings—it’s about building something India needs and making sure India uses more AI and adopts it. Singh pointed out that Indians already power much of the AI world. “[Consider] Perplexity, OpenAI—30–40% of them (the workforce) are Indians. So it’s not like India can’t do it. We just need time and support from the ecosystem.”

“The same thing is happening now that happened with cloud and silicon. We don’t want to repeat that. We want to make it work for India—by Indian developers,” Khatri said.

The post What Keeps Team Krutrim Up at Night appeared first on Analytics India Magazine.