info
- handle
- Ophuscado
- outfit
- Ophuscado LLC
- alias
- 圣黑客
- since
- 2003, a first website at twelve
- origin
- born in Spain, raised globally
- works
- remote first, on site worldwide
- speaks
- English and Spanish, fluently
- reads
- French, Italian and Portuguese
- learning
- Chinese
- member
- ACM, the Association for Computing Machinery
- status
- freelance, available full time
- nfo
- ophuscado.nfo, or
curl ophuscado.com
In a corner of the internet whose address I have no wish to recall, there lived not long ago one of those hackers who keep a shell open on the rack, a keyboard worn smooth, a lean server and a crawler for coursing. From the age of twelve he wanted to know, of every machine that crossed his path, how it worked and why; and what with little sleep and much reading of manuals, source code and other people's mailing lists, his brain dried up, and he resolved to become the most knowledgeable person about computers on the face of the earth.
That hacker is me, and I am not there yet. A first website in 2003, the first hacks in 2005, paid client work by 2007; then a studio in Spain, years in Singapore and China as a lead developer, a mail company co-founded, and now Ophuscado LLC. What the years left in my hands: shell scripting, PHP and Python for the web and the tooling; C, Rust and assembly for drivers and anything close to the metal; Linux and BSD servers set up, automated and kept alive; systems broken into, then locked; payments wired up; language models run on local GPUs. Lately, the Apple M2 Ultra under Linux, worked out from macOS's own tables and traces: PCIe tunnelled over USB4, the first on that chip; its Neural Engine running, also the first, now on by default upstream; and its GPU measured against Metal until the gap for language models had names and numbers, with long-context generation up to 4.6x faster.
“A jack of all trades is a master of none,” people say, and stop there. The line goes on: “but oftentimes better than a master of one.”
contact
- phone
- +1 (307) 222-3643
- pgp
- key on request, until it is published here
engagement
- contract
- freelance, through Ophuscado LLC
- hours
- full time, overlapping the client's time zone
- location
- remote; on site worldwide when required
- billing
- fixed price, hourly or monthly retainer
Mail for postbox@ophuscado.com can be encrypted to this key. Check the fingerprint against the one your mail app shows.
Write in English or Spanish. For work on an existing system, include your current setup, the change you need and the deadline. Long-term engagements and new, experimental or research-grade projects are equally welcome.
open source
Linux on Mac and PC hardware at the quality of the vendor's own system. Current platform: Mac Studio, M2 Ultra. Each record links its evidence.
- Apple Neural Engine Apple's Neural Engine running under Linux on the M2 Ultra, a first on record. The upstream overlay crashed the boot: on this chip the engine's memory mappers sit in a power domain (ANE_SYS) that must be switched on first, as macOS's power tables show. Fixed and merged the same day; on my test results the chip was enabled by default upstream the next day. proof: pull request 278 · pull request 297 · findings
- Thunderbolt 3
Thunderbolt 4
USB4 PCIe tunnelled over USB4 and Thunderbolt 4 on the M2 Ultra under Linux, also a first on record: the tunnel controller driver ported, its DMA offset found in macOS's tables, the cable speed fixed. The same kernel drives two 24-bay Thunderbolt 3 enclosures at 40 Gb/s, unchanged. proof: findings · kernel · kernel tree - Storage server
594 TB All 33 disks of 18 TB moved from Apple's HFS+ to XFS in place, every copy verified before its source was formatted. Now one 504 TB mergerfs pool behind four SnapRAID parity disks, with idle disks spun down. proof: findings · storage - llama.cpp
Vulkan backend Twelve patches for local AI on the Mac's GPU under Linux, measured against Metal on the same machine: prompts +12–23 %, long-context generation up to 4.6x faster, 9-token batches 3x faster so speculative decoding pays (a code edit from 30 to 55–59 tokens/s), a Metal-style shader for up to +7 % generation, an abort fixed. The rest of the gap, measured and reported upstream: each dependent GPU launch costs 4 µs here against 2.3 µs on Metal; not fusion, not memory. proof: issue 10982 · findings and table · patches - Magic Keyboard
Magic Mouse
Magic Trackpad Apple's keyboard, mouse and trackpad made to behave as on macOS, matched against its own measurements: the same pointer speed, key repeat and scrolling, no scrolling from a finger resting on the mouse (a driver patch), Fn shortcuts for windows, and the Bluetooth pairings carried over. proof: findings · desktop - KDE Plasma
desktop A desktop tuned after macOS: the WhiteSur look, window handling and shortcuts, Apple's SF fonts from Apple's own downloads, macOS's colour profile for the LG G5 OLED, and a task-switcher bug traced to libplasma and fixed. proof: findings · desktop - OhLED Themes
for VS Code 142 VS Code themes for OLED screens, one colour on true black, and 11,000+ installs. proof: Marketplace · source - linux-lab
Mac Studio lab Where all of it lives, GPL-2.0-or-later: each change as a patch series on a pinned upstream commit, the benchmarks, a public macOS reference dataset, and one script that rebuilds the setup from a fresh install. github.com/Ophuscado/linux-lab
…and many more adventures that I can no longer call to mind, and which, to tell the truth, your grace would not much care to hear.
projects
Built and run by me. How they work stays behind the curtain until each one ships.
- ForoHackers A semi-private forum for hackers, makers and the sharp-witted, with encrypted groups and a chat that forgets what you told it. forohackers.com
- HackedSystem classified until launch hackedsystem.com
- Intraneto A web directory for people tired of search results, curated by hand. intraneto.com
- Liverado Email that is really yours, on your own domain, with a straight answer about what is kept. A Liverado Pte. Ltd. service I co-founded. liverado.com
- Technofeudal classified until launch technofeudal.com
services
-
Software engineering
Web applications and APIs, CLI and desktop tools, kernel drivers and GPU shaders, from high-level frameworks down to assembly.
-
Infrastructure and networking
Linux and BSD servers for mail, web, DNS, file sharing and directories; routing, firewalls and VPNs of every common protocol, plus custom tunnels; on dedicated hardware or in the cloud.
-
Data, search and media
Databases and search, scraping, parsing and OCR pipelines, big-data processing, compression, visualization, and audio and video encoding and live streaming.
-
Reliability and operations
Infrastructure as code, self-verifying deployments, encrypted backups with tested restores, monitoring and incident response.
-
Security
Offensive and defensive: red teaming and adversary emulation, penetration testing, exploit development and reverse engineering; hardening, firewalls and intrusion detection; radio and hardware attacks; encryption and privacy by design.
-
Linux kernel and driver development
Kernel drivers, device trees and desktop integration for hardware without upstream support, built from reverse-engineered vendor interfaces. Current platform: Apple M2 Ultra.
-
Local AI and GPU compute
On-premises LLM inference: GPU backend tuning and profiling, speculative decoding, quantised models, and adversarial testing of models and agents; no data leaves your network.
-
Payments
Card and cryptocurrency checkouts, subscriptions, invoicing and reconciliation.
-
Technical advisory
Architecture reviews, technical due diligence and build-or-buy analysis for companies and investors.