// Romanian Brain Bee - Neuroscience Competition

Lead Technical Expert - Romania

2008 - present

Developed and maintained the online contest platform for thousands of Romanian students taking part in the Romanian stages of the Brain Bee neuroscience competition.

Visit RBB ↗ Brain Bee Global ↗

Latest videos

YouTube ↗

I gave my game's NPC a language model | Building nanoMMO #25
Watch on YouTube ↗
MMO zones and portals in C++ with raylib | Building nanoMMO #22
Watch on YouTube ↗
MMO items in C++ with protobuf text format | Building nanoMMO #26
Watch on YouTube ↗
Pretraining a GPT on a laptop in 10 minutes | Minimum Viable GPT #2
Watch on YouTube ↗
Fine-tuning a GPT to follow instructions | Minimum Viable GPT #3
Watch on YouTube ↗
Where a GPT's time goes, profiled | Minimum Viable GPT #6
Watch on YouTube ↗
Build a GPT from scratch: 14.6M parameters | Minimum Viable GPT #1
Watch on YouTube ↗
A KV cache makes a GPT write 4× faster | Minimum Viable GPT #4
Watch on YouTube ↗
A GPT in the browser, a quarter the size | Minimum Viable GPT #5
Watch on YouTube ↗
DPO: a GPT learns from preferences | Minimum Viable GPT #7
Watch on YouTube ↗
PPO: a GPT learns from its own stories | Minimum Viable GPT #8
Watch on YouTube ↗
A BPE tokenizer in C++, 116× faster | Minimum Viable GPT #11
Watch on YouTube ↗
RoPE: a GPT that reads past its window | Minimum Viable GPT #10
Watch on YouTube ↗
LoRA: fine-tune 2% of a GPT's weights | Minimum Viable GPT #9
Watch on YouTube ↗
Flash attention: 12 GB down to 48 MB | Minimum Viable GPT #13
Watch on YouTube ↗

Latest writing

Substack ↗

Training a 14M GPT to Generate Children Stories in 20 Minutes: Pretraining, SFT, and Direct Preference Optimization

End to end implementation, parameter and architecture analysis, and light evaluation of the checkpoints

Read on Substack ↗

Could Actor Systems Be All We Need for Agentic AI Workflows

Are We Reinventing the Wheel of Actor Systems as We Develop Agentic Orchestration for AI?

Read on Substack ↗

Is Computer Science a Worthwhile Pursuit Today?

As of 2026, It Will Most Certainly Endure the Test of Time, Albeit With a Raised Bar of Expectations

Read on Substack ↗

Latest videos

I gave my game's NPC a language model | Building nanoMMO #25

MMO zones and portals in C++ with raylib | Building nanoMMO #22

MMO items in C++ with protobuf text format | Building nanoMMO #26

Pretraining a GPT on a laptop in 10 minutes | Minimum Viable GPT #2

Fine-tuning a GPT to follow instructions | Minimum Viable GPT #3

Where a GPT's time goes, profiled | Minimum Viable GPT #6

Build a GPT from scratch: 14.6M parameters | Minimum Viable GPT #1

A KV cache makes a GPT write 4× faster | Minimum Viable GPT #4

A GPT in the browser, a quarter the size | Minimum Viable GPT #5

DPO: a GPT learns from preferences | Minimum Viable GPT #7

PPO: a GPT learns from its own stories | Minimum Viable GPT #8

A BPE tokenizer in C++, 116× faster | Minimum Viable GPT #11

RoPE: a GPT that reads past its window | Minimum Viable GPT #10

LoRA: fine-tune 2% of a GPT's weights | Minimum Viable GPT #9

Flash attention: 12 GB down to 48 MB | Minimum Viable GPT #13

Latest writing

Training a 14M GPT to Generate Children Stories in 20 Minutes: Pretraining, SFT, and Direct Preference Optimization

Could Actor Systems Be All We Need for Agentic AI Workflows

Is Computer Science a Worthwhile Pursuit Today?