The Turiya Labs Manifesto

While writing this in 2026, AI is the new buzz word everywhere, everyone wants everything AI. Every frontier AI lab is rushing to build the next best model which tops the benchmarks. Realistically everyone knows the competition is between China’s AI labs and America’s AI labs, while the American labs are keeping its smartest models proprietary despite having ‘Open’ in their name, China is open sourcing its smartest model signalling to the world that if they want intelligence they should go to China.

In India people at the top of their respective fields, who have led IT companies are saying that India doesn’t need to develop its own model and we should use models created by other labs and build products efficiently using those models. I won’t name those people, but everyone knows who they are. The lack of vision these so called ‘industry experts’ have is daunting. What we have instead is 10 min delivery apps, get a maid in 10 mins app, son of the richest man in India starting an ice-cream brand, people buying electronics from China in bulk for cheap and selling it in India calling it a home-grown brand then after the brand is profitable switch to getting the parts from China and just ‘assemble’ the parts in India and call it ‘Made in India’ with an asterisk mark. Don’t get me wrong I know and agree that to make any brand or any business it takes a shit ton of effort and perseverance and the 10-min delivery app technology and the way it works and the amount of jobs its providing is also very fascinating and the people who have made all these brands and technologies are innovators who have made it so convenient for us to live.

Even after saying all this, we have to agree that we don’t have startups in deep tech fields, fields such as defense, artificial intelligence, quantum computing, biotechnology, aerospace and so many more. The reason for that is firstly, the incentives in India are just not worth it. I mean even if I am passionate about creating something in India for Indians I would get zero upfront support from the government, yes there are specific schemes made for this but we all know how much those schemes work. Not only do I have zero support but I also have hindrances caused by people in the government itself (bureaucrats). If this wasn’t enough, Indian VCs either don’t want to invest in deep tech startups because it’s too risky or even if they want to, they just don’t have the capital that a deep tech startup would need to make quality stuff. Hundreds of billions of dollars have been poured into the AI labs in America and the highest investment in India for an AI lab has been $2 billion as of writing this.

So technically and by all means the odds are stacked up against us, but the good thing is that the odds have always been stacked against us and we as a country have beaten those odds and have come out on top. An example of this is the ‘Mangalyaan Mission’ (Mars Orbiter Mission) which was done on a budget of $74 million by the brilliant people at ISRO while the movie ‘The Martian’ starring Matt Damon was made on the budget of $108 million. We have the talent, we just need the intent. We need more indigenous deep tech startups and we need them now.

Turiya Labs is our way of contributing to this effort. While all the ‘big’ AI labs are chasing the biggest models to show the highest intelligence and racing towards what they call Artificial General Intelligence (AGI), we at Turiya Labs think that rather than making huge models which should be able to do everything, why not make small models designed to do specific tasks but the model does that task really well and ideally does it locally on any device. Task specific, small, efficient and private should be the goal for these models. So rather than Large Language Models (LLMs) we create Small Language Models (SLMs).

India is one of the largest smartphone markets in the world. For millions of people the smartphone is their first and often only computer. If AI is going to become part of life it cannot remain something that only runs on expensive GPUs in distant data centers. It has to run where people are. Small Language Models make that possible.

A designed SLM can execute on consumer hardware, edge devices and increasingly, directly on smartphones. That means latency, reduced infrastructure costs, better privacy and the ability to work even when connectivity is limited. Importantly specialization beats generalization in many real-world applications. A farmer asking about crop diseases does not need a model of writing Hollywood screenplays. A healthcare worker documenting information does not need a model that has memorized the entire internet. A lawyer researching case law does not need a model optimized for creative fiction. They need a model that deeply understands one domain and performs that task well. That is where task-specific SLMs excel.

For a country as diverse as India, with dozens of major languages, unique legal systems, healthcare practices, government schemes, educational curricula and countless local workflows, it is unrealistic to expect one giant model to serve every need equally well.

Instead we envision an ecosystem of models:

• Models built for agriculture.
• Models built for healthcare.
• Models built for law.
• Models built for education.
• Models built for manufacturing.
• Models built for services.

Each one is efficient to run on everyday devices. Each one trained for the people who actually use it. We do not believe intelligence has to be massive. We believe it has to be useful.

We know we can’t catch up with the frontier AI labs overnight. Training a state-of-the-art foundation model needs billions of dollars, huge computing power, and years of research. We can’t do it right now. What we do have is determination.

So we start with what we can do. We fine-tune small open models on practical problems for India. We test them. We enhance them. And we release them freely on Hugging Face for everyone to use them, assess them and let us know where they need improvement.

Every model takes us forward. Every release teaches us something. Every feedback helps us make the next model better. We don’t aim for the biggest models; we aim for the most valuable ones.

Perhaps some day in the future, Turiya Labs will be able to train foundation models from scratch. Perhaps some day in the future, we will have models that rival the best in the world. But before we get there, we have to deserve it.

This is how Turiya Labs starts.