Research focus
Nazarbayev intellectual schoolsShymkent
I build TPU-native Pallas kernels and train hybrid language models (GDN-2 + MLA). I created a 3D WebGL Observatory to visualize neural network internals. My current work focuses on efficient sequence modeling, numerical audits, and making AI more interpretable.
Machine Learning
LLMs
Efficient Architectures
AI Interpretability
Collaboration interests
Public collaboration signals submitted through this research profile.
Education
Nazarbayev intellectual schools
Latest publications
A preview of approved public publications linked to this researcher.
- Open
Atomic-GDN2: TPU-Native Kernels and Hierarchical Delta Routing for Hybrid Language Models
Zenodo (CERN European Organization for Nuclear Research) - 2026 - OpenAlex00 OpenAlex citations - Open
Atomic Ops: Fused Gated DeltaNet-2 Kernels for TPU v5e in JAX/Pallas
Zenodo (CERN European Organization for Nuclear Research) - 2026 - OpenAlex00 OpenAlex citations
Regional connection
DSML member