Computer Engineer | Systems Architecture | Bare-Metal Problem Solver

I am a Computer Engineering graduate and a current student of the Master's in High Performance Computing at the Universidade da Coruña (UDC), focused on low-level architecture, systems programming, and high-performance computing. I enjoy bridging the gap between hardware and software, from reverse-engineering automotive ECUs and compiling C parsers, to managing bare-metal hyper-compute clusters and rebuilding internal combustion engines.

Currently working as a Core Software Developer at Ágata Technology (Emetel Group), building smart multisystem integration platforms. Alongside my professional work, I am expanding my focus towards Embedded Systems, Electronic Warfare, Defense Tech, and high-performance systems programming (C/C++, Rust).

Still private while I work on them: hetfx, my Master's thesis, a complete FX correlator spanning an FPGA, GPUs over NVLink and InfiniBand, and multi-node runs on FinisTerrae III; and kv260-ias2rf, an RF front-end adapter board for the Kria KV260.

My technical background extends beyond standard software development. I approach engineering as a continuous puzzle across multiple disciplines:

• Systems Architecture & AI Infrastructure: Designing and maintaining self-hosted homelab environments. I manage a custom bare-metal Gigabyte T181-G20 server (Dual Xeon Platinum, 4x Nvidia Tesla V100 SXM2 via NVLink) deploying local LLMs (Open-WebUI, vLLM...) and a Docker swarm (Gitea, Minio...). Alongside it, an AMD Kria KV260 (K26 SOM, Zynq UltraScale+ ZU5EV) as the FPGA/SDR bench. Both are driven from the terminal by tools I wrote for the purpose — redfishctl, a Rust TUI that controls and monitors the server's BMC over Redfish (sensors, charts, power, event log) without ipmitool or the BMC web interface.
  • Hardware Hacks: Built a remote PSU trigger bridging PS_ON to GND using a Raspberry Pi GPIO and a MOSFET via Home Assistant. Modified the MegaRAC SP-X BMC JSON configurations to optimize fan curves for idle/load acoustics without thermal throttling.
• Low-Level OS Troubleshooting: Experienced in disaster recovery techniques. Examples include using sethc.exe backdoor exploitation for NT AUTHORITY\SYSTEM privilege escalation to rebuild a corrupted WinSxS directory, and forcing Tesla V100s into WDDM mode on Windows Server 2025, which NVIDIA does not expose for these cards at all — nvidia-smi offers only TCC — by writing the display driver keys directly in the registry, and hot-swapping back without a reboot. That unlocked the GPUs for WSL2 and for graphical workloads (Blender, games) while compute tooling such as LM Studio still detects them under TCC.
• Automotive Mechanics & ECU Engineering: Practical knowledge of engine mechanics (rebuilding engines, timing belt replacements, and resurrecting seized motors). Working knowledge of ECU calibration parameters (WinOLS, GDS2, SOI advanced timing, torque capping for gearbox protection) and of the diagnostic layer beneath them — reverse-engineering an adapter's undocumented protocol down to its CRC, and reading a factory tool's own 89 MB vehicle database by decompiling the loader that reads it. This led to my sh705x-emu and opendash.
• Hardware & Fabrication: Hands-on experience with 3D printing (Resin/FDM), micro-soldering, PCB assembly, and drone piloting.

📜 Academic Excellence & Endorsements

Throughout my degree (GPA: 8.7/10), I have achieved a record of excellence, receiving 7 Honors (Matrículas de Honor) in computer science and engineering subjects:

• Core Systems & Logic: Computer Structure, Programming Paradigms, and Language Processing (Flex/Bison).
• Data & Infrastructure: Databases, Computer Networks, and Information Retrieval.
• Research & Development: My Bachelor's Thesis (Framework for Data Capture and Synchronization for Brain-Computer Interfaces).

I have been endorsed by 6 University Professors and PhDs, five of them specifically for the Vulcanus in Japan international program. These recommendations, along with my official transcript, highlight my technical rigor, ability to solve low-level architectural puzzles, and dedication to high-complexity projects.

🔗 View Transcript & Recommendation Letters Here

• sh705x-emu: An emulator for Denso engine ECUs on the Renesas SH-2E, written from scratch in Rust, that runs a car's own firmware unmodified against modelled peripherals and a simulated engine — the flash idles at 750 rpm on the torque it asks for, as the car does, faster than real time. Traces every calibration-map lookup live, decoded from its own descriptor and checked against the firmware's own result, with an egui desktop GUI and a Kotlin/Compose Android app over the same core.
• scanner64: A 64-channel real-time DDC bank in Verilog for the AMD Kria KV260's programmable logic: one sample per cycle, fixed latency, dropped samples impossible by construction. Verified in hardware and benchmarked bit-exact against C, CUDA and Xeon builds — 11x a Tesla V100 and 40x a dual Xeon per watt.
• glidedesk: A GPU-accelerated remote desktop for Linux, in Rust: each user gets a headless GNOME Wayland session composited on the GPU, with no monitor or dummy plug, and the desktop resizes with the client window. Hardware encoding (NVENC, VA-API), served at once over standard RDP so mstsc works with nothing installed, over its own low-latency protocol (video over UDP with Reed-Solomon FEC, so a lost packet never stalls the picture), and over WebRTC to any browser.
• DiskLens: Disk health (S.M.A.R.T.) for Linux and Windows, in Rust, with CrystalDiskInfo's evaluation ported line by line, more than 40 SSD vendor and controller tables included. One engine behind a terminal UI, a CLI, a desktop GUI, and a web interface with a JSON API and Prometheus metrics. Checked against CrystalDiskInfo on four disks: 268 of 269 values match.
• NeuroSync-Framework: A synchronized pipeline for real-time EEG neurophysiological signal processing. Implements multiprocessing for isolated CPU-intensive tasks (FastICA) alongside a multi-interface architecture (GUI, CLI, FastAPI).
• Rustify: A music player for Android, Windows, Linux, macOS, iOS and the web: one native Rust core under one Compose Multiplatform interface. The core is reached over JNI or a C ABI and brings its own audio engine and DSP effects where there is no ExoPlayer, while rustify-server puts it behind an HTTP API and serves the whole app to any browser as WebAssembly. Also has Jam sessions over MQTT and Google Drive sync that merges across devices instead of overwriting.
• opendash: An Android diagnostic app for the Scanmatik SM3, built on an undocumented Wi-Fi protocol recovered from packet captures and confirmed against a live vehicle. Identifying the frame signature as a CRC turned replaying captures into constructing messages — real CAN filters, live data at ~100 frames/s, and an ELM327 bridge so any existing OBD app can drive the device.
• Fabric2D: A procedural 2D engine built on pygame with no commercial framework underneath. Features multithreaded Jump Point Search (JPS) pathfinding, graph-based procedural generation, and the native FMOD audio engine driven through its Python bindings (in collaboration with roiniti and Mateo-RR).

• Languages: C, Rust, Python, Jython, Julia, Java, OCaml, SQL, Kotlin, Clingo/Telingo, Bash/PowerShell (currently expanding into C++ & Verilog/SystemVerilog).
• Systems & Low-Level: Concurrency (threads, mutex/locks, multiprocessing), C-FFI and JNI, asynchronous and socket I/O, fixed-point DSP, binary formats and wire protocols, bare-metal deployment.
• Reverse Engineering: Ghidra, Wireshark and packet capture, CFR/JADX decompilation, CAN/ISO-TP and J2534 diagnostics, WinOLS/GDS2.
• Hardware & FPGA: Vivado/Vitis, XSim, KiCad, AMD Kria KV260 (Zynq UltraScale+), micro-soldering.
• Infrastructure & Services: Docker & Colima, Kubernetes, Kafka, MongoDB, Neo4j, OpenSearch, Valkey/Redis, Testcontainers, Nginx, Wireguard, Prometheus, MinIO, Redfish/BMC, Windows Server 2025, Ubuntu Server.
• How I Work: Git and GitHub Actions with CI gates; empirical debugging — measure first, change one thing at a time; the measurements behind a claim published with it, and writing down what is not verified as carefully as what is.

I tend to approach most challenges as logical puzzles, but I also know when to completely unplug:

• 🎻 Music: Violinist (9 years of professional conservatory training).
• 🇯🇵 Languages & Culture: Self-taught Japanese (memorized ~2,000 kanjis by visually deconstructing them into radical puzzles) and long-time anime enthusiast. Native Spanish/Galician, C1 English, basic German (A1), and currently exploring Russian.
• 🏹 Focus: Traditional archery, mechanical restoration, and spatial puzzles (e.g., Megaminx).
• 🌿 Outdoors: Hiking and exploring fervenzas (traditional Galician waterfalls) and the deep green natural landscapes of my homeland to disconnect from the grid.

"If a system has logic, it can be understood. If it can be understood, it can be optimized." — Personal Engineering Motto
"逃げたら一つ、進めば二つ" (If you run, you gain one; if you move forward, you gain two.) — Mobile Suit Gundam: The Witch from Mercury
"ちりも積もれば山となる" (Even dust, when piled up, becomes a mountain.) — Japanese Proverb