Research & Development

R&D at Dvyamsha

Exploring boundary-pushing architectures and lightweight frameworks for next-generation intelligence.

Physics-Informed ML

Microcosm: Library for Physics-Informed Neural Networks

A specialized library designed for modeling complex physical domains by embedding physical laws, conservation principles, and domain constraints directly into neural network architectures.

LLM Efficiency & Architecture

Focus-Retention Architecture for LLMs

Novel focus-retention attention mechanisms enabling ultra-low memory footprint, fast context retention, and optimized local inference performance on edge and commodity hardware.