Alan Kelly
Software Engineer at Citadel Securities
Zürich Metropolitan Area Switzerland
Summary
🤩
Rockstar🎓
Top SchoolAlan Kelly is a software engineer with 12 years of deep expertise in CPU-focused AI acceleration and high-performance inference, currently based in Zürich. He led XNNPACK development at Google—enabling TFLite, PyTorch and ONNX delegations—and wrote optimized 4- and 8-bit quantized matrix-multiplication kernels that powered Gemini Nano on CPU across Arm and Intel architectures. His work spans kernel-level assembly and C, runtime reshaping that boosted Gemini’s performance by ~20%, and major operator and memory-arena optimizations that cut memory or latency by orders of magnitude. Prior roles include delivering real-time event-camera inference at Prophesee and leading ML/HPC R&D at NumScale, where he built FFTs and compression tools competitive with industry libraries. An active contributor to XNNPACK, TensorFlow and MediaPipe, he combines rigorous academic training (PhD) with a proven track record of squeezing production ML workloads into constrained hardware.
12 years of coding experience
15 years of employment as a software developer
MSc High Performance Computing, MSc High Performance Computing at The University of EdinburghBachelor of Engineering (B.Eng.), Bachelor of Engineering (B.Eng.) at University of GalwayGaelcholaisteDoctor of Philosophy - PhD Engineering, Doctor of Philosophy - PhD Engineering at University of Glasgow
English, Irish, French
Stackoverflow
11reputation
101reached
0answers
1question
Github Skills (38)
unit-testing10
c-language10
machine-learning10
inference10
xpack10
tensorflow-lite10
deep-learning10
tensorflow10
cprogramming-language10
optimization10
test-automation10
performance-monitor9
performance-analytics9
performance-measurement9
performance-monitoring9
Programming languages (7)
Github contributions (5)
google/XNNPACK
Nov 2021 - Jan 2023
High-efficiency floating-point neural network inference operators for mobile, server, and Web
Role in this project:
Backend Engineer & Test Automation Engineer
Backend Engineer & Test Automation EngineerContributions:135 reviews, 117 commits, 6 PRs in 1 year 2 months
Contributions summary:Alan primarily contributed to the XNNPACK library by fixing a sanitization error and improving unit tests. Their work involved modifying test code to prevent coverage tests from timing out, and adding dimension normalization. The user also added several new microkernels, showing a focus on optimizing the library's performance. The contributions show a combined emphasis on both backend operations and testing.
multithreadingsimdtensorflowcpufloating
tensorflow/tensorflow
Nov 2021 - Jan 2023
An Open Source Machine Learning Framework for Everyone
Role in this project:ML Engineer & Software Engineer
ML Engineer & Software EngineerContributions:54 reviews, 70 commits, 2 PRs in 1 year 2 months
Contributions summary:Alan's contributions primarily involve optimizing and enhancing the TensorFlow Lite framework, focusing on improving its performance and compatibility. Their work includes modifying and adding new functionality to the kernels, such as adding support for new operators like "Indy LSTM". They also contributed to optimizations for existing operators, such as improvements to the batch matrix multiply operator, and for XNNPACK. Further, they were responsible for testing a variety of models.
pythondata-sciencedeep-learningmlmachine-learning
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.