Raul Perez Pelaez

Research Scientist @ Universidad Autónoma de Madrid

About

I accelerate compute-intensive workloads by combining deep CUDA expertise with modern software engineering practices. Over the past decade, I have built and optimized GPU libraries that support drug discovery, fluid dynamics, and deep‑learning frameworks used by thousands. What I do • Design CUDA/C++ libraries from scratch. e.g., UAMMD for complex‑fluid simulation. • Maintain and distribute GPU‑heavy packages through channels such as conda-forge. • Engineer robust CI/CD with tools such as GitHub Actions, CMake, and Conda so every commit ships with tests, docs, and binaries in a reproducible manner (GitOps). • Build GPU‑accelerated web apps (React/Astro + WebGPU/WebAssembly) and advanced OpenGL visualizers (compute shaders, SSAO, PBR). • Containerize and orchestrate workloads via Docker / Compose / Kubernetes. How I work – Emphasis on automated testing, documentation, and reproducible builds. – Cross‑disciplinary: comfortable jumping from low‑level CUDA kernels to TypeScript front‑ends or LLM tooling. – Community-minded: conference organiser, university lecturer, and PhD mentor. I thrive where physics, ML, and high‑performance engineering intersect. Seeking roles that enable me to push GPU hardware to its limits, whether that’s cores in an exascale cluster or GPUs in the cloud. Let’s connect if you need someone to turn “could be faster” into “is 50× faster.” I excel at writing scalable, testable, documented software following healthy community standards.

Country

Spain

City

Madrid

Industry

Computer Software

Skill

High Performance Computing (HPC), Slurm Workload Manager, GitOps, complex fluids, Conference Organization, Event Planning, Project Management, DevOps, SYCL, conda, WebAssembly, WebGPU, Docker compose, Docker, jax, CMake, Nsight Compute, Continuous Integration and Continuous Delivery (CI/CD), GitHub, Triton

Experience

Research Scientist

Universidad Autónoma de Madrid

2024-10 - Present · 1 yr 11 mos

Madrid, Comunidad de Madrid, España

• Developed several GPU-ready open source tools for high-performance scientific computing (C++/CUDA with Python wrappers). • Set up GitHub actions with custom runners for automatic testing and benchmarking in a local GPU cluster. • Developed new massively parallel algorithms for biosensor modelling, published in scientific journals. • Wrote a high-performance web application (WebAssembly via C++ with Emscripten, WebGPU, Astro/React) for a digital twin product. • Developed and implemented algorithms for research in liquid memory storage. • Mentored two PhD students.

Adjunct Professor

IE University

2025 - Present · 1 yr

Lecturer (Adjunct) in the Computer Science and Applied Math degrees. Teaching courses such as: • Computer Programming 2 (compsci): C++ • Computer Programming 2 (applied math): Advanced Python • Computer Programming 1 (compsci): Advanced C

Conference Organizer

Freelance

2025 - Present · 1 yr

Founder and chair: First international DoNEV conference. • Conceived and organized an international meeting (CECAM flagship event) on non-equilibrium dynamics (60 + attendees from 10 countries). https://www.cecam.org/workshop-details/dynamics-of-non-equilibrium-variables-multiscale-multiphysics-applications-of-fluctuating-hydrodynamics-1408

Research Director

Universitat Pompeu Fabra

2023-1 - 2024-10 · 1 yr 10 mos

Barcelona, Cataluña, España

Provide core GPU libraries and tooling to research groups, industry, and OSS projects, mainly for AI-assisted molecular modeling and drug-discovery. Actions • Implemented custom torch.compile-ready, Autograd-enabled, CUDA ops in TorchMD‑Net; profiled & refactored neighbor search path. • Wrote and maintained conda‑forge recipes for OpenMM, TorchMD‑Net, OpenMM‑Torch, CUTLASS, pyg‑lib, NNPOPS, and more. • Integrated CUDA Graphs into OpenMM‑Torch for smoother ML‑augmented MD runs. • Built CI/CD matrices (GitHub Actions) with GPU, CPU, and cross‑platform wheels, plus auto‑generated docs. • Ported Neural Network Potential ML models to JAX, TinyGrad, and TensorFlow. • Wrote a Slack App integrating a local LLM model acting as an assistant in the internal workspace. The backend LLM model offered functionality such as tool usage, RAG, model hot-swap, paper summarization, and more. The LLM ran in an in-house GPU computing cluster, capable of hosting sub-60b parameter models (Llama family, Mistral,...). • Finetuned LLM models for custom workflows in drug discovery. Impact • TorchMD‑Net inference time reduced from hours to minutes (50 × speed‑up). • conda‑forge packages adopted by multiple bio‑physics labs; eased deployment on HPC clusters.

Postdoctoral Researcher

Universidad Autónoma de Madrid

2022-9 - 2023-11 · 1 yr 3 mos

Research software engineer and postdoctoral researcher. Also took care of the local GPU computing cluster. • Managed a SLURM HPC cluster with 600 CPU cores and 40 NVIDIA GPUs scattered around 15 computing nodes.

Research assistant

New York University

2021-9 - 2022-9 · 1 yr 1 mo

Nueva York, Estados Unidos

Predoctoral researcher in the area of fluid physics and complex systems

Universidad Autónoma de Madrid

2015-6 - 2021-9 · 6 yrs 4 mos

Madrid y alrededores, España

I specialize in the development of software for complex fluid simulations, implementing parallel code that leverages the GPU and CPU. • Conceptualized and developed Universally Adaptable Multiscale Molecular Dynamics (UAMMD), a complex fluid simulation toolkit with state of the art performance, written in CUDA/C++.