● About the Lab & Author

Abhishek Shinde

AI Developer & Independent Researcher

Oropis is where research and writing on efficient inference, low-bit LLMs, and hardware-aligned systems is published.

Mission

Short version. Building and studying models that fit closer to the hardware they run on.

Longer version. Oropis works in public on low-bit language models, shift-oriented inference, and systems questions around them. The aim is direct: make the useful parts clear enough to be verified, benchmarked, and carried forward.

Email is open at norizzabhii@gmail.com.

Timeline

Jul 2026TetraNet — power-of-two quaternary weight paper & kernels.
2025–2026AI development, Triton kernels & low-bit quantization experiments.
2024First decoder language-model architecture experiments.

Key Project

Research Release

TetraNet

Power-of-two quaternary weights for small decoder-only language models. Eliminates FP multipliers while reclaiming the unused 4th state of 2-bit storage registers.

quantizationpytorchc++triton

Publications

Aug 13, 2026

Blog

Abhishek Shinde

TetraNet

Emergence Is a Property of Opponents: Self-Play and the Curricula That Games Build

NNUE: The 22-Megabyte Neural Network That Saved Chess Engines

The Era of 1-bit LLMs: Three Symbols, 1.58 Bits, Zero Multiplies

TetraNet: Power-of-Two Quaternary Weights for Small Decoder-Only Language Models