Gregory Parisotto Reichert
I build for people who push
computers too far.
Twelve years building Razor in Brazil taught me where machines actually break. Not in benchmarks. In the boring failures nobody demos: a thermal wall at hour six, a driver that lies, a queue backing up while someone waits to finish real work.
Now I build software that gets closer to the machine, so computers stop being so dumb about themselves.
CurrentlyBuilding local AI and computer-intelligence systems, closer to the operating system.
01 / Work
All work- 01
Built Razor
Founder & CEO · 2014 to 2025High-performance computers for people whose work could not wait. Engineers, architects, researchers, studios. I learned that performance is not a benchmark. It is whether someone finishes the job.
- 200,000+ people advised
- 5,000+ customers
- US$24M+ in sales
- Petrobras, Samsung, SAP, Globo
- 02
Studying the machine layer
OngoingThe place where hardware, software, and human frustration meet. Telemetry, diagnostics, drivers, thermals, the failures that never make keynote slides. Most software floats too far above it.
- 03
Building local intelligent systems
Now · quietSoftware that gets closer to the machine. Local AI and computer intelligence, so machines understand themselves and adapt to the person instead of the other way around.
02 / Writing
All writingEfficiency needs to become a product feature again
July 17, 2026I caught Google Chrome peaking at 7.13 GB of RAM on my machine.
Hardware Always Wins: Why Apple Proves the Future Belongs to the Physical
September 2, 2025Software gets the headlines. Hardware gets the world.
On the Brink of Ruin: How Jensen Huang Nearly Lost Everything Before Building Nvidia
July 29, 2025The most valuable chip company on earth was, more than once, thirty days from dead.
Why Quantum Computing Will Not Revolutionize Technology (At Least Not Yet)
January 6, 2025Comparing quantum to classical computing is like comparing an airplane to a submarine.
03 / Notes
All notesMost dashboards are confessions. They show you everything the system failed to understand on its own.
The closer you get to the machine, the less patience you have for noise.
Performance problems are rarely technical at first. Usually they are decisions nobody wanted to make, showing up later as latency.
Brazil teaches you to build without the fantasy of infinite money, time, or bandwidth. It is a harder teacher than most accelerators, and a better one.