• 🔭 Current Focus: Hardware-aware LLM optimization, GPU orchestration, and custom Triton kernels.
• 🚀 Nebius Academy: Mastering high-performance infrastructure for the Israel National AI Supercomputer.
• 🛠️ Experience: 25+ years in low-level R&D, from compilers to high-frequency trading engines.

An open-source VS Code extension designed to improve code documentation by rendering Markdown directly within code comments. It helps maintainers and developers visualize rich text, lists, and links without leaving the editor.

• AI Compute: Triton, CUDA Kernels, GPU Memory Optimization (HBM3/SRAM), Inference Scaling.
• Systems Architecture: High-Performance Computing (HPC), Distributed Clusters, C++, Go, Rust.
• MLOps: GPU Orchestration, Docker/K8s for AI, Latency-Critical Backend Systems.
• Legacy Expertise: Compiler Design (Lex/Yacc), R&D, Real-time Search, and IoT Data Streams.

I am documenting my deep-dive into AI Performance Engineering here:

• AI Performance Engineering Repo - Benchmarks, Triton kernel optimizations, and GPU scaling experiments.

Founder & CTO | vkhey! Mar 2025 – Present

• Architecting high-performance multimodal RAG systems (text, video, audio) for enterprise scale.
• Bridging 25 years of systems engineering into the "plumbing" of the agentic AI era.

Principal Developer | Labguru May 2023 – Mar 2025

• Engineered a first-of-its-kind cross-server communication platform for distributed lab environments.
• Led technical implementation for ISO 27001, GDPR, and SOC 2 compliance in high-security environments.

Principal Developer | Kando Jan 2021 – Mar 2023

• Tech lead for the National COVID-19 monitoring project, managing massive IoT data streams.
• Optimized data-processing algorithms and high-performance dashboards for national-scale infrastructure.

Principal R&D Dev | SeekingAlpha Jul 2015 – Jul 2020

• Built multi-million dollar ad products and high-throughput analytics pipelines for 20M+ monthly users.

Note: While I am a polyglot (Python, Ruby, Node.js, Go, C#, C++), I specialize in System Design where the choice of tool is dictated by hardware constraints and performance requirements. Currently spending my days in Triton and C++ to squeeze every TFLOP out of the H100.

• LinkedIn: linkedin.com/in/valhk
• Website: vkhey.com
• Email: [email protected]

This profile and current R&D projects are powered by real-time web grounding via the Brave Search API.