Alexey Radul
Principal Software Engineer at Positron AI
Belmont, Massachusetts, United States
Summary
🤩
Rockstar🎓
Top SchoolAlexey Radul is a Principal Software Engineer with 18 years of experience building compilers, probabilistic programming systems, and high-performance ML infrastructure. Currently at Positron AI, he leads core work on transformer execution, continuous batching, and KV caching for accelerator-backed workloads. His background includes developing TensorFlow Probability and the research language Dex at Google, with notable open-source contributions to Dex and JAX—improving type systems, IR passes, and ragged/tensor batching primitives. Trained at MIT (PhD/MEng/BS), he blends deep formal expertise in programming-language design, static analysis, and automatic differentiation with pragmatic engineering for production systems. Known for inventing elegant mechanisms to tame complexity, he favors Haskell and obsessively refactors compilers and shape/batching code for correctness and performance.
18 years of coding experience
14 years of employment as a software developer
PhD, Computer Science, PhD, Computer Science at Massachusetts Institute of Technology
Russian, English
Stackoverflow
66reputation
1kreached
1answer
0questions
Github Skills (22)
batching10
python10
operation10
tensorrt10
numpy10
type-system10
compiler-design10
tensorflow10
batch10
tensor10
jax10
code-optimization9
intermediate-code9
linear-algebra9
intermediate-language9
Programming languages (8)
Github contributions (5)
google-research/dex-lang
Apr 2019 - Jan 2023
Research language for array processing in the Haskell/ML family
Role in this project:
Back-end Developer
Back-end DeveloperContributions:320 reviews, 369 commits, 324 PRs in 3 years 9 months
Contributions summary:Alexey primarily focused on implementing and refining the compiler's internal workings, with commits showing changes to the type system, optimization passes, and the inliner. The contributions involved core aspects of the compiler's intermediate representation (IR), including adding support for features like dependent pairs, improving the precision of effect annotations, and refactoring the structures involved with the occurrence analysis. The user also implemented the refactoring of the parser to emit a safer names representation.
arrayhomotopy-type-theoryautomatic-differentiationhaskellarray-processing
jax-ml/jax
Nov 2020 - Nov 2023
Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more
Role in this project:Back-end Developer
Back-end DeveloperContributions:17 reviews, 5 PRs, 19 comments in 3 years
Contributions summary:Alexey primarily contributes to the JAX library by implementing and refactoring core functionalities related to batching and array manipulation. Their work focuses on enhancing the `RaggedAxis` representation for handling dynamic shapes and integrating it with existing primitives like `dot_general`, `broadcast_in_dim`, and `einsum`. This includes modifying batching rules, shape computations, and implementing slicing operations, and improving performance for ragged tensor operations. The user also addresses code style issues by fixing flake8 errors.
pytorchpythonjitautomatic-differentiationgpu
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.