Who we are
An AI lab that builds and ships
Sankhya is both a research lab and a product company. We build models, and we prove them through SensAI — our live teaching platform serving real learners and institutions.
About Sankhya AI Labs
An AI lab that builds small models and proves them in production
We are not building AI to be big. We are building AI to be good at specific tasks — memory, speech, and learning — and we prove every model through SensAI, our live product. We are both a research lab and a product company.
The name
Sankhya: structure, clarity, and disciplined thinking
In the Samkhya tradition, understanding comes from careful enumeration and analysis — not from noise or scale. That's how we think about AI. Not larger models. Not more features. Better focus. Clearer purpose. Smaller, sharper tools.
Why India
India's constraints make us build better models
Multiple languages. Mixed devices. Weak connectivity. Dense classrooms. These are not problems to solve later — they are the design constraints that force our models to be smaller, faster, and more resilient. If our models work in India, they work anywhere.
Who we are
A lab that builds models and a company that ships products
We don't just publish papers. We build models, integrate them into a live product (SensAI), and prove them with real learners and institutions. Every model we build is tested in production.
What we believe
Small models can outperform large ones at specific tasks
The industry's obsession with scale is not the only path. Task-specific, compact models — trained well, deployed efficiently — can deliver better results where it matters.
Where we're going
Independent AI for India's learning systems
Our long-term direction is a full stack of small, owned models that can run independently — powering personalized teaching in every classroom, regardless of connectivity or infrastructure.
Principle 1
Build small, build focused
We don't build models to be big. We build them to be good at a specific task — memory, speech recognition, voice generation, edge deployment.
Principle 2
India is the design constraint, not an afterthought
Language diversity, classroom density, device variability, and connectivity gaps are core product inputs — not edge cases we'll handle later.
Principle 3
Prove it through product
Every model we build is tested in SensAI with real learners. We don't publish benchmarks in isolation. We ship and measure in production.
Principle 4
Own what matters. Use what helps.
We use external APIs where they help us move faster. We build our own models where quality, privacy, cost, and offline capability demand it.