Diogo Silva
Portrait of Diogo Silva

I work on making software fast on supercomputers.

I'm an MSc student in Advanced Computing at the University of Minho in Braga, Portugal, focused on parallel programming, distributed systems, and performance optimization. I recently interned at Bosch, building AI-assisted security tooling.

Now
Master's thesis on disaggregated memory for heterogeneous systems

Selected work

Disaggregated Memory in Heterogeneous Systems

2026–now

Master's thesis, in progress

A memory-disaggregation framework for heterogeneous HPC and data-center environments. It investigates remote-memory access, communication overhead, data placement, and performance-aware memory management.

C++, Python, Bash, MPI

Grade 20/20

How practical is LLM inference without a GPU? A benchmarking harness for the Deucalion supercomputer that measures quantization, thread scaling, context size, request concurrency, BLAS variants, and multi-node RPC with llama.cpp.

C/C++, Python, Bash, llama.cpp, GGUF, SLURM · with Gabriel Paiva and Tomás Pereira

1.51 TFLOPS on Fujitsu A64FX · Grade 18.3/20

A custom double-precision matrix multiply using BLIS-style cache blocking and packing, SVE-512 vectorization, and CMG-aware OpenMP, measured on a 4096 × 4096 matrix with 48 threads. It also includes CUDA kernels for the NVIDIA A100 and a blocked LU factorization.

C, OpenMP, SVE, CUDA, SLURM · with Tomás Pereira

Grade 18/20

A plasma simulation code tuned across OpenMP, hybrid MPI + OpenMP, and CUDA, then compared on the Fujitsu A64FX, AMD EPYC 7742, and NVIDIA A100 nodes of Deucalion.

C, OpenMP, MPI, CUDA · with Tomás Pereira

A scientific-visualization reconstruction of the tsunami: far-field propagation from the Horseshoe Fault with adaptive mesh refinement in GeoClaw, plus representative near-field flow studies in OpenFOAM.

GeoClaw, OpenFOAM, ParaView · with Frederico Silva and Tomás Pereira

Grade 18/20

A graphics engine with a procedural model generator and XML-defined scenes, ending in a textured, animated solar system with hierarchical transformations.

C++, OpenGL, GLSL · team of four

The engine rendering a textured solar system, with planets on elliptical orbits around the sun.

Tools I built

A local web GUI for browsing, searching, previewing, and transferring files over SSH. I built it for Deucalion, and it works with any host in ~/.ssh/config. It only needs the Python standard library and the system's ssh and scp.

Python, JavaScript

The ssh-fetcher interface showing a remote file list, a file preview, and a console.

HPC Forge

2025

An interactive CLI that generates optimized Makefiles and SLURM job scripts, with presets for clusters like Deucalion. Install it from PyPI.

Python

The HPC Forge terminal menu with options to generate a Makefile or a SLURM job script.

Experience

Software Developer Intern, Bosch Portugal

Jul–Sep 2026

Developed a TypeScript VS Code extension for real-time security vulnerability detection, and integrated generative AI APIs for context-aware remediation and automated fix suggestions.

EUMaster4HPC Summer School

Jul 2026

Explored emerging HPC and AI topics with students, researchers, and HPC professionals from across Europe.

Education

MSc in Advanced Computing

2025–now

University of Minho, Braga

BSc in Computer Science

2021–25

University of Minho, Braga

Skills

Programming
C, C++, Python, TypeScript, Java, SQL
HPC and parallel
MPI, OpenMP, CUDA, SYCL, SLURM, performance optimization
Systems and tools
Linux, Bash, Git, Docker, Apache Spark, Milvus, MySQL
Languages
Portuguese (native), English (professional), Spanish (basic)
Activities
EUMaster4HPC Summer School, EPICURE Hackathon, ENEI, English B2 (Babelium)

Contact

diogo.coelho.silva@gmail.com

© 2026 Diogo Silva