Pedro Bello-Maldonado
Systems Engineer @Luma
Signup · Get unlimited contacts
WORK HISTORY
Systems Engineer @Luma
Miami-Fort Lauderdale, FL, US
Modernized multimodal GenAI training and inference infrastructure from Slurm to Kubernetes achieving full performance parity for both communication and computation- Migrated our codebase to run on Flyte workflows allowing users to manage infrastructure and runtime plans directly from their Python scripts- Enabled priority scheduling, job preemption, and automated failure recovery through a combined resource management architecture involving Kueue, Flyte, and Kubernetes- Achieved over 90% cluster utilization on GPUs accross multiple compute providers- Created a CLI for easy interaction with Kubernetes and Flyte resources that allows a seamless transition from HPC in Slurm to HPC in Kubernetes
EDUCATION
Florida International University
Master's Degree, Electrical and Electronics Engineering
Florida International University
Bachelor's Degree, Electrical and Electronics Engineering
University of Illinois Urbana-Champaign
Master’s Degree, Computer Science
University of Illinois Urbana-Champaign
Doctor of Philosophy (Ph.D.), Computer Science
ABOUT PEDRO BELLO-MALDONADO
I\'m a Systems Engineer at Luma AI with focus on high-performance computing (HPC), artificial intelligence (AI), machine learning (ML), and cloud computing. I work implementing efficient computing infrastructure with Slurm, Kubernetes, and VMs to enable distributed training, inference, and physics simulations in AI/ML applications. I also focus on the performance optimization of GPU kernels to improve the evaluation of machine learning models. For my PhD, I worked on the solution of partial differential equations (PDEs) on HPC systems with special interest in high-order methods and iterative solvers and preconditioners, as well as GPU computing on distributed systems. These methods are widely applicable to many other fields including optimization, quantitative analysis, fintech, and more. Similarly, implementing these methods on HPC systems involves knowledge of different computing languages and understanding of computing resources to achieve the best performance possible.
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.