Professional and open-source experience

My work connects generative AI systems, numerical kernels and computer architecture. My earlier experience spans RISC-V vector programming, ARM SIMD, multicore execution, instrumentation and control systems.

AI Performance Engineer · Generative AI (R&D)

Current role

My current R&D work focuses on the software and execution performance of generative AI systems.

  • Develop and optimize low-level compute kernels and customize runtimes for AI and machine learning workloads on hardware accelerators.
  • Build agentic engineering systems to automate performance optimization tasks.
  • Benchmark, profile and analyze AI systems, and prepare technical reports on the results.
  • Design, implement, test and maintain software for AI products and internal engineering platforms.
  • Contribute to technical investigations, delivery support, documentation, code review and knowledge transfer.

My current toolchain includes vLLM, CUDA, C++ and Clang.

Open-source developer · rv-sparse

Started June 2026 · Remote · Completed collaboration

I contributed to rv-sparse through a program of The Linux Foundation, sponsored by RISC-V International. Project repository and contributor credits.

  • Contributed to the development of an open-source sparse linear algebra library for the RISC-V Vector Extension (RVV), focusing on vectorized execution, memory access and portability across RVV-capable processors.
  • Applied vectorization, loop restructuring and data-layout choices to hardware-aware sparse kernel implementations.
  • Worked toward a research-oriented evaluation covering correctness validation, performance benchmarking and documentation of design decisions.

Research Engineer · Universitat Politècnica de València (UPV)

January 2021 – April 2024 · Valencia, Spain

  • Developed and optimized BLAS/BLIS-style GEMM, MatMul and GEMM-based convolution kernels for deep learning inference on low-power ARM and RISC-V processors.
  • Tuned compute loops using ARM NEON and RVV intrinsics, loop unrolling, instruction-level optimization and data-layout transformations.
  • Parallelized kernels with OpenMP and applied tiling and scheduling that account for cache and memory hierarchies.
  • Integrated and validated kernels in RTOS environments under memory and real-time constraints; built Python/Bash harnesses for correctness checks, benchmarking and performance regression tracking.

Related evidence: matrix multiplication project · technical deep dive.

Development Intern · Institute for Molecular Recognition and Technological Development (IDM)

May 2019 – January 2020 · Valencia, Spain

  • Modeled systems in MATLAB and implemented validated models in C/C++ on AVR microcontrollers.
  • Developed and validated electronic instrumentation for analog and digital signal acquisition.
  • Implemented digital components in Verilog, verified their functional behavior and performed static timing analysis against design constraints.

Application Engineer · KEA Electronics

January 2017 – June 2018 · Ecuador

  • Integrated sensors into UAV platforms using I2C, SPI, CAN and UART; developed embedded components and Python automation for data collection.
  • Developed orthophoto processing workflows with Python and QGIS, and supported server administration for processing and storage.

R&D Intern and Assistant Researcher · UTPL

August 2015 – June 2017 · Ecuador

  • Designed and validated optimal control functions for robotic-hand phalange positioning and grip-force regulation.
  • Implemented control functions on ARM/AVR embedded targets and analyzed experimental results to evaluate stability and performance.

Research projects at UPV

I contributed as a Research Assistant & Engineer to work supported by European and Spanish research programmes. The periods below are the official project lifetimes; my UPV position is dated separately above.

eFlows4HPC · Horizon 2020 / EuroHPC

Project period: January 2021 – February 2024 · Grant 955558

eFlows4HPC developed methods and software for workflows combining HPC, data analytics and machine learning, with UPV among its partners. My related work covered BLIS-like matrix multiplication and performance analysis for low-power processors; the corresponding papers acknowledge this funding.

European Commission project record · GEMM paper and acknowledgements · Performance analysis paper.

Low-power machine learning acceleration · TED2021-129334B-I00

Project period: December 2022 – August 2025 · AEI / EU NextGenerationEU–PRTR

The project, officially titled Aceleración de algoritmos de aprendizaje automático en microcontroladores y microprocesadores de consumo ultra-reducido, supported research on efficient machine learning for constrained processors. My related publications address parallel GEMM-based convolution and communication-avoiding fusion on GAP8. The 2024 parallel-convolution paper credits my implementation of GEMM and execution of the experiments.

UPV project record · Parallel convolution: funding and contributions · Convolution-fusion publication.

Energy-aware HPC for deep learning · PID2020-113656RB-C22

Project period: September 2021 – February 2025 · Spanish State Research Agency (AEI)

This UPV research project concerns energy-aware high-performance computing and communications, with applications to computational deep learning. My related work examined numerical kernels, memory traffic and execution on low-power processors. The project received Spanish national research funding.

UPV project record · Funding acknowledgement.

Research evidence

Matrix computation on RISC-V · 2022

The 2022 work examines BLIS-like integer GEMM on the GAP8 Fabric Controller and the performance of numerical workloads at the edge. The accompanying case study separates the published implementation from educational examples and proposed optimizations.

A BLIS-like matrix multiplication for machine learning in the RISC-V ISA-based GAP8 processor · Research poster (PDF).

Deep learning convolution algorithms · 2024

The 2024 publications address parallel GEMM-based convolution on multicore RISC-V processors and communication-avoiding convolution fusion on GAP8, connecting algorithm design, parallel execution and data movement.

Parallel GEMM-based convolution for deep learning on multicore RISC-V processors · Communication-Avoiding Fusion of GEMM-Based Convolutions for Deep Learning in the RISC-V GAP8 MCU.

Embedded sensing and neural processing

The master’s thesis record describes an embedded system for assessing meat maturation using impedance sensing and neural processing. Earlier publications cover control, sensing and automation applications.

Institutional thesis record · Complete publication list.

Doctoral research · 2025

The doctoral thesis brings together matrix computation and deep learning convolution algorithms for low-power RISC-V processors. The doctorate was completed in 2025.

My doctoral research was supported by a Santiago Grisolía fellowship from Generalitat Valenciana, acknowledged in the research publications.

Doctoral thesis record.

Selected technical training

  • PATC@BSC: Parallel Programming (MPI/OpenMP); Performance Analysis & Tools.
  • LRZ/NVIDIA DLI: Deep Learning Week, covering Transformers and multi-GPU workloads.