Àlex Batlle Casellas

alex.batlle01@gmail.com | +34 620 93 76 28 | GitHub | LinkedIn | CV

Hi, I’m Àlex — a Machine Learning Engineer at Qualcomm Europe and a graduate student in Mathematics at UPC, based near Barcelona.

At Qualcomm I work on ML infrastructure and LLM research, with a focus on large-scale distributed training. I led multi-node LLM pretraining on clusters of up to 60×8 H100 nodes, working on RDMA/RoCE-based communication, NCCL collective tuning, and performance portability across network fabrics. This work resulted in two peer-reviewed publications at SC Workshops ‘25, one of which received the Best Paper Award at INDIS. I have also contributed to research on efficient training techniques (quantization-aware training, knowledge distillation at HPC scale), wireless network modeling, and LLM reasoning.

On the academic side, I am pursuing an M.S. in Mathematics at FME-UPC, with interests in Graph Theory, Computational Complexity, Algorithmic Game Theory, and HPC for AI. I previously completed a double Bachelor’s in Mathematics and Data Science & Engineering at CFIS-UPC — a highly selective excellence center (~35 students admitted per year). My Bachelor’s thesis, written during a research visit to TU Delft, focused on autonomous robot exploration using Graph Neural Networks and Reinforcement Learning (available online).

My side projects and course notes can be found on my GitHub profile.