Skip to main content

Research

Efficient AI for the languages and hardware people actually have

We are an early-stage lab exploring efficient sequence models, native Tamil language understanding, and vision-language intelligence. We build products from what we learn and open-source what we can.

Current Focus

Approach

Constraints first

We design for the hardware and budgets people actually have. If it only runs on an A100, it is a demo, not a product.

Build to learn

We build working systems, not just papers. Implementation surfaces the hard questions that theory hides.

Open by default

Architecture research and reusable tools go open-source. Product-specific work stays proprietary.

Publications

Papers and open-source

Working toward our first publications. Code and weights will be released under Apache 2.0.

In preparation

Our first papers are in the grounding phase. Preprints will go on arXiv and code on GitHub as work matures.

Working on similar problems?

We are open to collaboration, especially with researchers in the Indian AI community.