Skip to content

DEV Community

Yuri Pocepaev

Software engineer at Neuroprem. LLM quantization, efficient inference, retrieval and agent memory. Practical benchmarks and runtime fixes. Less hype. More engineering.

Joined on Sep 27, 2026 https://neuroprem.com

Work

Software engineer and lead developer at Neuroprem

Oct 5

When Offloading Stalls Inference: Queue Capacity, CUDA Event Ownership, and Misleading Health Checks

#architecture #backend #debugging #performance

9 min read

Sep 29

One Model, Many Roles: Specializing LLM Inference Without Training More Models

#ai #llm #architecture

12 min read

Sep 29

Prevent Context Leakage in Multi-Tenant LLM Systems

#security #privacy

6 min read

Sep 28

Smaller Context, Recoverable History: Inside an Agent Memory Handoff

#agents #ai #architecture #llm

8 min read

Sep 27

How I Took GLiClass FP8 from 59 ms to 16 ms on an RTX 4050

#machinelearning

8 min read

Sep 27

Building an NVFP4 KV Cache for a Hybrid Qwen Model

#machinelearning

10 min read

Sep 27

Running Qwen Flash-Next NVFP4 in vLLM: PLE Loading, B12x Fixes, and Stable Inference

#machinelearning

13 min read

loading...