-
01
From VB6 to Linux
My first language was VB6, back in elementary school, then VB.NET, mostly for very
silly purposes. I broke Windows often enough that I eventually tried Linux, and picked
up a DistroWatch-roaming habit: install, break, hop, repeat. Bash scripting crept in
slowly, and somewhere along the way I watched
Jadi's LPIC-1 course.
Little helpers like
AA-Script and
HyperFast
came out of that era: less about elegance, more about refusing to click the same thing twice.
-
02
The robotics phase
Then came robots. Oxygen was a helmet-mounted safety device for cavers:
MQ gas sensors watching temperature, humidity, carbon monoxide, carbon dioxide and methane,
so you'd know the air was turning before you did. I built a line tracer, and a lightweight
junior football-playing robot that I took to IranOpen. Years later at university, the habit
resurfaced as an AC remote: an ESP8266 talking MQTT to control air-conditioning units.
-
03
Libraries, Rust and weird numbers
The scripts grew into Python libraries like
Maryam and
RadixHopper,
Python led to Rust and Go, and somewhere along the way I fell for math, compilers and
anything with a weird number system. Hyperoperators are still my favourite rabbit hole.
-
04
The spark
In 2019, GPT-2 wrote paragraphs that sounded almost human. In 2021, DALL·E drew an armchair
shaped like an avocado. Interesting, I thought. Clever.
Then, at university, Mahdi
started talking about large language models. He was better at this than me, and he was
sure we'd need more breakthroughs, new architectures, before anything like real
intelligence. But something in me went quiet. Better architectures would come, sure. They
always do. I just had faith, the kind of certainty you only get a few times in life, that
this was already enough. Enough to reach AGI. Enough, eventually, to reach ASI. Everything
I've done since has been, one way or another, chasing that moment.
-
05
First steps: Sirius and BERT
My first ML project was
Sirius,
a Persian-language assistant bot with a body. To be honest, Ali and Mahdi, together with
Koorosh Moslemi,
did most of the BERT work: intent detection and named-entity recognition. My job was
mainly leading the part where hardware met software, getting the models, the APIs and
the physical bot to actually talk to each other.
On hardware we were, by far, ahead of everyone else. And it still failed: the Raspberry Pi's SD card gave up
on us at the worst possible moment, and Ali erased the wrong hard drive. Models are only
part of the story; plumbing, backups and a little luck are the rest.
-
06
LLMs, and explaining them
At university I went deep on LLMs, from attention to quantization to optimizers, and
found I enjoy explaining them almost as much as building with them. That turned into
talks, lectures and the Randrange podcast, all a little further up this page.
-
07
Today
I joined DuoAnalytics as an AI engineer and now lead its engineering team, while doing
my MSc in AI at the University of Guilan. My days are mostly architecture, reviews, and
helping the team ship. My nights go to reading papers and side projects, as long as the
mornings don't stretch too far, which they often do :)
-
∞
A few names along the way
Aryan Horizon, Alireza Mohammadnezhad, and AARMN the Limitless
(Aryan AliReza MohammadNejad),
and counting. The surname is really Mohammadne(j/zh)ad: technically it's zh,
but j is what most people write, so you'll find me under both.
To infinite, to infinite.
