AI Infrastructure & Compiler Engineer

Compiler and Systems Engineer specializing in MLIR-based code generation pipelines, cross-platform neural network acceleration, and hardware-oriented graph transformations. Experienced in progressive IR lowering (Linalg → GPU → SPIR-V), extensible inference engines, and mathematical acceleration techniques (Winograd, GEMM, RVV).

• Compiler Engineering & IRs: LLVM, MLIR (Linalg, GPU, SPIR-V, MemRef, SCF, Arith), Dialect Conversion, TableGen, progressive lowering, lit, FileCheck.
• Heterogeneous & GPU Computing: Vulkan 1.3 Compute, SPIR-V binary shader generation, Shader ABI & Descriptor Management, OpenCL, SIMT execution models.
• Languages & Core Systems: C/C++ (17/20), Python, CMake (modular & plugin architectures), Ninja, Linux, Docker, GTest.
• Architecture & Acceleration: RISC-V Vector (RVV), GEMM tiling & packing, Winograd (F(2,3)), im2col, shape-generic codegen, memory hierarchy optimization.
• ML Frameworks & Toolkits: OpenVINO, PyTorch, TensorFlow.

• OpenVINO (Intel / Open Source):

  • Enabled neural network inference optimization for RISC-V hardware architectures.
  • Designed and tuned high-performance GEMM kernels and native RVV vector microkernels, delivering up to a 100x speedup on compute-bound layers.
  • Researched subgraph tokenization and graph conversion interfaces (MarkPattern, GraphConverter) within the intel_gpu plugin.

• Adept — Automatic Differentiation & Math Lowering Engine:

  • Engineered a standalone C++ autograd and neural inference engine from scratch.
  • Formulated and implemented the Winograd algorithm for 3x3 convolutions, yielding a 2.25x – 3x inference acceleration compared to standard spatial convolution.

• Yandex ML Training:

  • Successfully completed tracks 1.0, 2.0, and 3.0. Currently deepening expertise in Reinforcement Learning (ML 4.0).

• University: UNN (Lobachevsky University), CS (FIIT), Department of High-Performance Computing and Systems Programming.
• Telegram: @idolnot4u
• LinkedIn: linkedin.com/in/strelkovkm
• GitVerse: gitverse.ru/strelkovkm
• Email: strelkovkm96@outlook.com