About

Applied computer science researcher specializing in machine learning compilers, tensor program optimization, and high-performance computing. My work has resulted in 12 peer-reviewed publications with over 290 citations. I am the lead author of the CGO 2025 Distinguished Paper Award and actively serve the scientific community on Artifact Evaluation Committees and review boards for top-tier venues like ASPLOS, CGO, and ACM TACO.

Honors & Awards

Distinguished Paper Award (Top 3%) International Symposium on Code Generation and Optimization (CGO) Awarded for the novelty and potential field impact of "Tensorize: Fast Synthesis of Tensor Programs from Legacy Code."	2025
Carl Zeiss Thesis Award Carl Zeiss AG Awarded for the most outstanding computer science master's thesis at universities in Dresden, Germany.	2021

Work Experience

Google Software Engineer Scaling PyTorch LLM training and maximizing hardware utilization across large scale TPU clusters. Developing novel extensions for Google's AI compiler infrastructure targeting PyTorch / TPU clusters.	Nov 2025 – current
Meta Research Scientist Intern Researched and developed a PyTorch extension for dynamic shapes in CUDAGraphs, achieving 4-10% speedups in LLM prefill and training.	Dec 2024 – Apr 2025
AMD Deep Learning Compiler Intern Prototyped MLIR kernel tiling and extended the AMD's AI engine compiler with shape inference.	Jun 2023 – Sep 2023

Education

University of Edinburgh PhD in Computer Science Thesis: Tensor Program Optimization through Program Synthesis	Feb 2022 – Oct 2025
Technische Universität Dresden Diplom (MSc) in Computer Science	Mar 2020

Professional Service

Journal Reviewer: ACM TACO (2025, 2026), Parallel Computing (2025, 2026)

Conference Reviewer: CGO (2022)

Artifact Evaluation Committees: ASPLOS (2026), CGO (2026), CASES (2025)

Software

Tensorize
Program synthesis based tensor compiler, achieving avg. 4.1x speedup by lifting programs to PyTorch, JAX and StableHLO. Uses novel top-down program synthesis algorithm that scales linearly on average. Code

MLIR Synth
MLIR program synthesizer, discovers equivalent programs in high-level MLIR dialects, like LinAlg and StableHLO. Code

PolyGym
Exploring valid loop transformations in polyhedral model with reinforcement learning, finding schedules with avg. 1.83x speedup over Polly/ISL. Code

Publications

STENSO: Tensor Program Superoptimization through Cost-Guided Symbolic Program Synthesis A. Brauckmann, A. Chaube, J. W. de Souza Magalhães, E. Polgreen, M. O'Boyle Paper, Code	CGO '26
Accelerating Sparse Algebra with Program Synthesis J. W. de Souza Magalhães, S. Hashemian, A. Brauckmann, J. Woodruff, E. Polgreen, M. O'Boyle Paper	CC '26
Tensorize: Fast Synthesis of Tensor Programs from Legacy Code using Symbolic Tracing, Sketching and Solving A. Brauckmann, L. Jaulmes, J. W. de Souza Magalhães, E. Polgreen, M. O'Boyle Distinguished Paper Award 🏆 Paper, Slides, Code	CGO '25
Guided Tensor Lifting Y. Li, J. W. de Souza Magalhães, A. Brauckmann, M. O'Boyle, E. Polgreen Paper, Code	PLDI '25
DFA-Net: A Compiler-Specific Neural Architecture for Robust Generalization in Data Flow Analyses A. Brauckmann, A. Faustino da Silva, G. Synnaeve, M. O'Boyle, J. Castrillon, H. Leather Paper, Code	CC '25
Guess, Measure & Edit: Using Lowering to Lift Tensor Code J. W. de Souza Magalhães, J. Woodruff, J. Armengol-Estapé, A. Brauckmann, L. Jaulmes, E. Polgreen, M. O'Boyle Paper, Code	PACT '25
MLIRSynth: Automatic, Retargetable Program Raising in Multi-Level IR using Program Synthesis A. Brauckmann, E. Polgreen, T. Grosser, M. O'Boyle Paper, Slides, Code	PACT '23
Rewriting History: Repurposing Domain-Specific CGRAs J. Woodruff, T. Koehler, A. Brauckmann, Chris Cummins, Sam Ainsworth, M. O'Boyle Paper, Code	arXiv '23
ExeBench: an ML-scale dataset of executable C functions J. Armengol-Estapé, J. Woodruff, A. Brauckmann, J. W. de Souza Magalhães, M. O'Boyle Paper, Code	MAPS '22
Polygym: Polyhedral Optimizations as an Environment for Reinforcement Learning A. Brauckmann, A. Goens, J. Castrillon Paper, Code	PACT '21
Compiler-based Graph Representations for Deep Learning Models of Code A. Brauckmann, A. Goens, S. Ertel, J. Castrillon Paper, Code	CC '20
ComPy-Learn: A Toolbox for Exploring Machine Learning Representations for Compilers A. Brauckmann, A. Goens, J. Castrillon Paper, Code	FDL '20
A Case Study on Machine Learning for Synthesizing Benchmarks A. Goens, A. Brauckmann, S. Ertel, C. Cummins, H. Leather, J. Castrillon Paper	MAPS '19

Skills

Programming	C, C++, Python
Frameworks	JAX, MLIR, PyTorch
Machine Learning	Transformer Models, Graph Neural Network Models, Reinforcement Learning, Performance Optimization
Other	Symbolic Execution, Program Synthesis, Linux, Git, Docker