Andrey Malyshev
Sr. Software Engineer at Deelvin Solutions Inc.
Paphos Municipality, Cyprus
Summary
🤩
RockstarAndrey Malyshev is a senior software engineer with over two decades of systems-level experience and more than a decade focused on deep learning compiler and deployment engineering. He designs and optimizes inference pipelines and compilers—most recently extending TVM for diverse hardware (Intel/Adreno/NVIDIA GPUs, Metal/OpenCL/CUDA, Intel/ARM CPUs, and Cortex‑M) and building an end-to-end quantized LLM serving stack with 8-bit and FP8 support. At Deelvin and previously at Intel he drove performance-critical work (OpenVINO, TVM) including low-batch throughput tuning, memory planning for OpenCL textures, and adding fast int8 X86 kernels, often turning research ideas into production-grade toolchains. He combines low-level binary analysis and instrumentation expertise with practical server and scheduling architecture design, enabling real-world speedups (for example up to 8x on Adreno GPUs). Active in the TVM open-source ecosystem, he has fixed subtle Metal vector race/accuracy bugs and added qnn operators and tuning improvements that materially raised compiler accuracy and performance. Based in Cyprus, Andrey pairs deep optimization skills with a pragmatic focus on delivering deployable, hardware-aware ML solutions.
11 years of coding experience
22 years of employment as a software developer
English
Stackoverflow
64reputation
3kreached
8answers
0questions
Github Skills (21)
performance-monitor10
machine-learning10
metal10
deeplearning-ai10
compiler-compiler10
performance-analysis10
deep-learning10
compiler10
c-language9
python9
cprogramming-language9
opencl8
computer-vision6
ionic-framework6
google-chrome6
Programming languages (3)
Github contributions (5)
apache/tvm
Apr 2021 - Jan 2023
Open deep learning compiler stack for cpu, gpu and specialized accelerators
Role in this project:
Backend Developer & Performance Engineer
Backend Developer & Performance EngineerContributions:174 reviews, 31 commits, 38 PRs in 1 year 9 months
Contributions summary:Andrey primarily contributed to improving the performance and accuracy of the TVM compiler stack, particularly in the context of Metal and quantized operations. Their work included fixing data race issues related to the Metal framework's vector data types and addressing recast issues in Relay operators. The user also added support for the qnn batch_matmul operator. Further contributions included optimizing the TVMC tuner for cases when uTVM is not enabled and enhancing X86 int8 support by adding fast SSE4/AVX2 instructions.
metalvulkancompilertensoropencl
elvin-n/dldt_tools
Apr 2019 - Nov 2022
Place for dldt based tools, samples and demos
Contributions:25 commits, 3 PRs, 36 pushes in 3 years 7 months
placetoolkitdldtpackinterfaces
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.