Product Labs🔬: Decoding Krutrim AI

In one of our previous editions, we tried to share some insights about the landscape of AI in India, touching upon some of the cool Indigenous LLMs being developed to comprehend Indic languages in addition to English. To commemorate the 78th Indian Independence Day, Spark & Trouble have decided to dive into one of the frontrunners in the Indian LLM space - Krutrim AI.

Branded as “India’s own AI”, Krutrim is a bold stride in India’s AI landscape. It can understand 22 Indian languages and generate content in 10 Indian languages.

What’s in it for you?

Krutrim AI is a prominent player in India's burgeoning AI ecosystem. Led by Bhavish Aggarwal, the entrepreneurial force behind Ola Cabs, Krutrim is focused on developing robust multilingual AI solutions. As one of the early entrants into the Indian AI market, it became India’s first AI Unicorn & has been instrumental in shaping the landscape and demonstrating the potential of AI for the country's diverse population.

Krutrim's assistant app is powered by a cutting-edge language model trained on a massive dataset of over two trillion tokens, with a strong emphasis on Indian languages. This makes it capable of understanding and generating intelligent responses in several Indic languages - including Marathi, Hindi, Bengali, Tamil, Kannada, Telugu, Odia, Gujarati, and Malayalam. You can interact with the app using your voice or text, and it can generate a variety of text formats, from simple answers to creative writing pieces.

Krutrim is the perfect example of a product achieving product market fit.

Krutrim achieves product-market fit by addressing a critical need for multilingual AI capabilities and offering a user-friendly platform that caters to diverse users, from businesses to individuals, across various industries and regions. Krutrim is also tailored to capture the nuances of India's rich heritage, which sets it apart from AI models predominantly trained on Western data.

While it may not yet be able to match up in terms of response quality to the AI big guns, like OpenAI ChatGPT, Google Gemini, Anthropic Claude, and Microsoft Copilot, there are multiple applications in the Indian landscape. Krutrim AI empowers businesses, content creators, students, and researchers alike.

Individuals

• Content Creation: Generate creative text formats, from poems to scripts, in multiple languages

• Personal Assistant: Get help with tasks like translation, summarization, and answering questions

• Students: For students and researchers, it's a vast knowledge repository accessible in their preferred language

Enterprises

Krutrim AI has a host of services targeted at enterprise customers

• Multilingual AI: Krutrim's core strength lies in understanding and generating text in multiple Indian languages. This is crucial for bridging the digital divide in India

• AI Cloud Platform: The platform offers AI computing infrastructure, foundational models, and open-source models for developers and enterprises to build their AI applications

• GPU-as-a-Service: Providing access to high-performance computing resources for AI training and development

• Foundational Models: Offering pre-trained AI models that can be fine-tuned for specific tasks and industries

• Location Services APIs: Providing location-based services and mapping solutions

Based on these services, here are a few industries where Krutrim AI could have large-scale applications

• Healthcare: By enabling communication between doctors and patients in local languages, Krutrim AI can improve access to healthcare services for millions in rural and remote areas.

• Governance: Krutrim AI can facilitate communication between government agencies and citizens in their preferred languages, leading to greater transparency and citizen engagement.

• Education: Educational materials and tools can be developed in local languages using Krutrim AI, making education more accessible and inclusive for diverse communities.

• Financial Services: Financial institutions can leverage Krutrim AI to provide financial literacy and access to financial products and services in local languages, promoting financial inclusion.

In fact, since charity begins at home, Ola’s CMO reported that all Ola group companies have already started utilizing Krutrim for a variety of internal tasks, including customer support, voice and chat, and customer sales calls.

Beyond LLMs… Into Krutrim’s AI Cloud Services

Another intriguing aspect of Ola’s Krutrim is that it goes beyond merely being an LLM-powered chatbot. The Krutrim Developer Platform offers AI Cloud Services such as APIs, foundational models, and infrastructure for individuals & businesses to build world-class products.

The highlight here is definitely the novel “Ola Maps” feature, which was introduced very recently as probably one of the first serious competitors to Google Maps. Ola Maps claims to deliver real-time, accurate navigation, uniquely designed for India's dynamic road conditions and ever-changing environments.

Krutrim AI has a unique position to empower India's SMEs. By offering multilingual capabilities, it breaks down a significant barrier for businesses struggling with English proficiency. With AI-driven automation for tasks like customer service, marketing, and operations, SMEs can focus on core competencies, increase efficiency, and compete effectively in the global marketplace. Krutrim AI is more than just a tool; it's a catalyst for growth and innovation for India's entrepreneurial spirit.

What’s the intrigue?

Sam Altman, the CEO of OpenAI, was in India in June of last year when he said that it is "totally hopeless to compete with us on training foundation models."

"You shouldn't try," Alman said at an event organized by the Economic Times. "It's your job to like try, anyway. I believe both of those things. I think it is pretty hopeless."

Krutrim AI represents a bold step towards challenging the notion that India is playing catch-up in the AI race. By focusing on multilingual capabilities and addressing the specific needs of the Indian market, Krutrim AI is carving a unique niche. While building a model on the scale of OpenAI's is undoubtedly ambitious, Krutrim AI demonstrates the potential for focused innovation.

However, building Indian LLMs is not without its challenges. The availability of high-quality training data in diverse Indian languages is a critical hurdle. Additionally, ensuring fairness and mitigating biases in these models requires meticulous attention. Krutrim AI’s success in navigating these challenges will be crucial in determining its long-term impact.

Whatcha Got There?!🫣

Buckle up, tech fam! Every week, our dynamic duo “Spark” ✨ & “Trouble”😉 share some seriously cool learning resources we stumbled upon.

While we celebrate the 78th Indpenedence Day in India, we couldn’t help but share this amazing tribute that AajTak AI created a year ago, by leveraging AI Tools to generate a version of our National Anthem featuring various freedom fighters & leaders.

This effort definitely gave us a sneak peek into the quality of cool stuff that’s possible with AI going forward. Do give it a watch 👇

You Asked 🙋‍♀️, We Answered ✔️

Question: We see that every day, hundreds of novel research papers related to AI are being published, demonstrating newer state-of-the-art approaches to solving problems. However, their actual infusion into products, to create actual user impact, seems like a different level of struggle altogether. Why is that the case?

Answer: The translation of AI research into impactful products is indeed a complex process & the gap between research breakthroughs and their practical implementation in products is indeed significant. This phenomenon is often referred to as the "research-to-practice gap" or "valley of death" for innovations.

There could be several reasons for this. Some of the most compelling ones are:

• Scalability challenges: Many research papers demonstrate effectiveness on small, controlled datasets or in specific conditions. Scaling these approaches to real-world, diverse, and large-scale applications often presents unforeseen challenges.

• Computational requirements: Some state-of-the-art models require extensive computational resources that may not be feasible or cost-effective for commercial products.

• Robustness and reliability: Research prototypes may not be sufficiently robust or reliable for real-world deployment where consistent performance is crucial.

• Regulatory and ethical considerations: Many AI applications, especially in sensitive areas like healthcare or finance, need to navigate complex regulatory landscapes and address ethical concerns before deployment.

• Business considerations: The potential return on investment for implementing cutting-edge AI techniques may not always be clear, making companies hesitant to invest heavily in unproven technologies.

• Skill gap: There's often a shortage of professionals who can effectively bridge the gap between academic research and practical implementation.

Bridging this gap effectively requires concerted efforts from researchers, engineers, business leaders, and policymakers. Strategies like creating more industry-academic partnerships, developing better frameworks for AI deployment, and focusing on interpretable and explainable AI can help accelerate the transition from research to impactful products.