Vlad Esaulov

Hi, I'm Vlad Esaulov

I'm a Ph.D. Candidate studying High-Performance Computing and Big Data

Understanding distributed systems and high-performance computing from the perspective of data optimization techniques.

I am a Ph.D. Candidate in Computer Science at Georgia State University working as a Research Assistant with the CS Department. My research focuses on building more efficient systems, data transfer frameworks for HPC applications, optimizing CUDA kernels for machine learning inference, and developing AI-powered security systems. I collaborate with Oak Ridge National Laboratory researchers on cutting-edge HPC applications.

I am actively seeking internship opportunities as a Software Engineer, Machine Learning Engineer, or Research Scientist to apply my expertise in high-performance computing, CUDA optimization, and distributed systems to real-world challenges.

News

Sep 2026 🎉 Paper "CephMaestro: A Safety-Gated AI Agent for Natural-Language Ceph Storage Management" accepted as a full paper at the IEEE ICPADS 2026, Tokyo, Japan
Apr 2026 🎉 Paper "JANUS: Resilient and Adaptive Data Transmission for Enabling Timely and Efficient Cross-Facility Scientific Workflows" accepted at the 35th ACM International Symposium on High-Performance Parallel and Distributed Computing (HPDC '26), Cleveland, OH, USA
Feb 2025 🚀 Successfully completed Multi-Precision Data Refactoring Framework research project, achieving 20% reduction in HPC data transmission time
Dec 2024 âš¡ Optimized GPU machine learning training speed by 15% through CUDA kernel optimizations and memory access improvements
Jul 2024 📄 Paper "Revisiting Erasure Codes: A Configuration Perspective" accepted at HotStorage 2024, ACM, Santa Clara, CA

Research Highlights

CephMaestro architecture
IEEE ICPADS 2026

AI Agents for Storage Management

CephMaestro: a safety-gated AI agent that manages Ceph clusters through natural language, with 112 typed tools, approval gates, dry-runs, and audit logging.

JANUS data refactoring and transmission pipeline
ACM HPDC '26

Cross-Facility Data Transmission

JANUS: resilient and adaptive data transmission for timely, efficient cross-facility scientific workflows. Refactors data hierarchically and protects it with erasure-coded fragments sent over wide-area networks.

Fault-injection testbed for erasure coding in distributed storage
ACM HotStorage '24

Erasure Coding in Distributed Storage

Revisiting Erasure Codes: how configuration affects erasure-code recovery in real storage systems. Configuration choices significantly change recovery time, and write amplification runs 32-72% above theory.

CUDA GPU Acceleration
Project

GPU Acceleration & CUDA

CUDA kernel optimization for machine learning workflows, improving GPU training speed by 15% through memory access pattern improvements.

GPT Architecture
Project

Large Language Model from Scratch

Built and trained a GPT-like LLM from the ground up using PyTorch, implementing attention mechanisms, pretraining, and instruction finetuning. Complete educational implementation mirroring foundational models like ChatGPT.