# Anvikshara

Intelligence worth building. An Indian frontier AI research initiative, exploring foundation models, compute efficiency and the science of safer intelligence.

> Formation stage. No released models, published research results or operating compute infrastructure are announced on this site.

## Our ambition

Built from India, with a global horizon. The next chapter of AI should have more authors.

Our ambition is to build an Indian frontier AI research company: one that develops its own scientific understanding, contributes original work, and treats safety as a research discipline.

Anvikshara is at the beginning. The immediate priority is finding a technical co-founder and shaping a focused, testable first research programme.

## Proposed research agenda

These are areas we intend to investigate. They describe our direction, not completed results.

1. Foundation models: what should intelligence be able to understand? Investigate how foundation models learn, reason and generalize. Begin with a bounded question, a reproducible evaluation, and a clear account of what the evidence supports.
2. Compute efficiency: how far can each unit of compute go? Explore the relationship between capability and the resources it requires. Study efficient training and inference, with measured trade-offs and reproducible comparisons.
3. AI safety: how do we know when to trust it? Study model behaviour, limitations and failure modes. Make evaluation and safety part of the research process from its first experiment.

## Research principles

1. Ask original questions. Choose problems that reward understanding. Make room for findings that challenge the starting assumption.
2. Let evidence lead. Distinguish an ambition from a result. Make methods, comparisons and limitations legible.
3. Make resources count. Start with a tractable experiment. Learn what matters before committing to scale.
4. Build with responsibility. Consider the consequences of capability alongside capability itself.

## The founding chapter

A research company begins with the right people. Anvikshara is seeking a technical co-founder with deep experience in machine learning research, foundation models or efficient compute, and the patience to build from first principles.

Meet the founder: https://callingbenjamin.com

Canonical page: https://anvikshara.com/
