The tech industry is witnessing rapid change after the introduction of ‘large language models’ (LLMs) like Chatgpt, Gemini, Llama, etc. Western tech giants dominate the field of LLMs, and companies like Open AI, Google, and Meta are the pioneers.

Recently, India has emerged as a significant player in LLM development. The country is simultaneously working toward technological advancement and cost-effectiveness, proving that LLMs can be developed with lower expenses and without compromising quality.

How did India succeed in doing so? Is it a robust infrastructure? Is it the vast pool of tech graduates? Or a combination of both in a conducive ecosystem for innovation?

The most visible factor that enabled India to develop LLMs at a low cost is its rapidly growing digital infrastructure, not to mention the huge investment the country has made in sectors like cloud computing, data centers, AI hardware, etc. Instead of using costly AWS or Google Cloud, companies like Tata Consultancy Service and Infosys use Indian cloud platforms and have created in-house AI frameworks. Besides, the country’s National Supercomputing Mission (NSM) is establishing low-cost, high-performance computing facilities.

On the other hand, prestigious institutions like the IITs and other top engineering colleges produce a steady stream of skilled engineers and computer scientists. The workforce is not limited to tech hubs like Bengaluru or Hyderabad. Talented engineers from small cities and towns contribute to cutting-edge AI research and development. These engineers are more affordable to hire compared to their Western counterparts.

To reduce development costs, Indian firms embraced open-source frameworks like TensorFlow, PyTorch, and Hugging Face. So, Indian developers can build on existing LLM architecture instead of starting from scratch. This saves time and drastically cuts down resource costs, enabling them to focus more on fine-tuning and optimizing models.

Also, Indian tech companies are participating in the global AI communities, allowing themselves to benefit from shared resources, open research, and collective intelligence. Developing LLMs independently is way more burdensome.

On the other hand, the government is proactively supporting AI research and development through funding and other initiatives, creating an environment where startups and companies can thrive. Collaboration between academia, industry, and government makes LLM development fast and cheap.
We all know how expensive training LLMs can be. You would need a vast dataset or access to them to train a model. Western companies pay hefty fees to acquire and process this critical component of LLM training.

Here comes an interesting benefit for India. The country has a vast and diverse population. People there speak dozens of languages and dialects (22 official languages, numerous dialects, and a heterogeneous terrain).

So, India offers a rich trove of localized data that can be harnessed to train language models. This allows Indian developers to create highly accurate and contextually relevant models without relying on expensive, curated datasets used in the West. Indian LLMs are not only more affordable but also more applicable to the local context.

Now, let us look at some thriving LLMs in India. BharatGPT is a project by CoRover (an AI startup under CEO Ankush Sabharwal) already being adopted in over 100 live implementations. Accessible in 22 languages by text and 14 languages by voice, BharatGPT is uniquely positioned to address the nuances of Indian languages.

Krutim is another major Indian LLM initiative that has utilized over two trillion tokens. Google-backed Project Vaani collaborates with 12 states and 80 districts to amass one of India’s largest dialectal datasets.

Sarvam AI’s OpenHathi 7B Hi has set benchmarks in translation accuracy across Indic languages. It is built on Meta’s Llama2-7B and was trained through bilingual modeling. In some bilingual evaluation metrics, OpenHathi outperformed models like GPT-3.5 and GPT-4.

Indian Railway’s IRCTC platform has implemented CoRover’s AskDisha chatbot to manage over 150,000 passenger queries daily with a remarkable 90% accuracy. Another notable adoption is by Indraprastha Gas, which launched a chatbot named Ask Maitri for customer support, reducing call volumes by over 35% within six weeks of launch.

Western developers often have access to more funding and resources from established tech giants. This, on the one hand, leads to groundbreaking advances and, on the other hand, causes environmental issues. Indian developers emphasize ‘jugaad,’ as in fostering innovative solutions instead of massive financial input.

Western companies have the advantages of global networks and collaboration, which are almost absent in India. However, Indian firms are tapping into global partnerships and leveraging their cost-effective development process to collaborate with international players.

Though Western nations are still ruling the AI landscape, India’s cost-effective development of LLMs signals that innovation is not limited to high-budget projects. More companies and startups are joining the AI race in India, and the country’s shift towards AI is now visible.

The author is an electrical and electronic engineering graduate and a regular writer for the tech industry.