Moisés Fernández
Research Engineer Manager at Synthesia
Murcia, Region of Murcia, Spain
Summary
🤩
Rockstar🎓
Top SchoolMoisés Fernández is a Research Engineer Manager with 7+ years of experience applying parallel computing and GPU expertise to large-scale scientific and ML problems, currently leading research engineering at Synthesia. He combines deep CUDA, OpenMP and MPI skills with practical ML and numerical optimization know-how—having optimized core components of MXNet and Gluon-NLP such as multi-tensor BERTAdam, FP16 inference and embedding-gradient GPU kernels. His background spans high-impact roles at NVIDIA and JPMorgan where he accelerated transformer training/inference and productionized TensorFlow-based pricing and portfolio frameworks. Trained as a neuroscientist (PhD, Oxford) and early CUDA developer for diffusion MRI and tractography, he uniquely bridges domain research and production ML systems, delivering up to 200x speedups on neuroimaging workloads. Based in Murcia, Spain, he blends hands-on kernel development with team leadership and a track record of contributions to widely used open-source deep learning projects.
7 years of coding experience
11 years of employment as a software developer
Colegio San BuenaventuraDoctor of Philosophy (PhD) Neuroscience, Doctor of Philosophy (PhD) Neuroscience at University of OxfordComputer Science Computer Systems Networking and Telecommunications, Computer Science Computer Systems Networking and Telecommunications at Université Claude Bernard Lyon 1Computer Engineering Area of expertise: Parallel Computing., Computer Engineering Area of expertise: Parallel Computing. at Universidad de Murcia
English, Spanish
Github Skills (26)
gpgpu10
python10
gpu-programming10
mxnet10
machine-learning10
gluon10
deep-learning10
gpu10
bert10
cuda10
nlp10
optimization10
testing9
operation9
c-language9
Programming languages (2)
Github contributions (5)
dmlc/gluon-nlp
Jan 2020 - Mar 2021
NLP made easy
Role in this project:
ML Engineer
ML EngineerContributions:5 reviews, 8 commits, 9 PRs in 1 year 2 months
Contributions summary:Moisés made significant contributions to optimizing the BERTAdam optimizer, including multi-tensor support and performance improvements. They added and refined tests for the optimizer to ensure its correctness. Further, the user implemented and integrated FP16 support for BERT QA inference, enhancing model efficiency. Their work also involved modifying the SQuAD script and associated modules to support mixed-precision operations.
nlpmxnetnatural-language-understandingnlgnatural-language-generation
apache/mxnet
Oct 2019 - Feb 2021
Lightweight, Portable, Flexible Distributed/Mobile Deep Learning with Dynamic, Mutation-aware Dataflow Dep Scheduler; for Python, R, Julia, Scala, Go, Javascript and more
Role in this project:ML Engineer
ML EngineerContributions:5 reviews, 18 commits, 23 PRs in 1 year 4 months
Contributions summary:Moisés focused on optimizing the performance of the embedding gradient computation within the MXNet deep learning framework. They implemented and optimized GPU kernels for embedding gradients, addressing lint issues and refining the code. They also contributed to the multi-tensor LAMB optimizer, incorporating multi-tensor sumSQ and optimizing GPU kernels. These changes demonstrate a focus on improving the efficiency and functionality of core deep learning operations.
pythonschedulerdataflowmutationdata-science
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.