Hey everyone, I'm Nikita.
I'm starting this blog as a home for my technical notes, side projects, and random deep dives. To be completely honest, I don't have a rigid roadmap for where this blog is heading—part of it is simply wanting a space for creative and technical self-expression.
A Quick Background
I’m a Ukrainian-Russian engineer by training, with around 4 years of experience in system administration. Over the last 1.5 years, I’ve been actively diving into Machine Learning, with a particular interest in cognitive science and reinforcement learning.
The Spark
Like many hobbies that turn into obsessions, it started with a single project: building and training a small-scale, AlphaZero-inspired model combining CNN + DQN + LSTM architectures.
It wasn't an overnight success. It took roughly three months of relentless tweaking—redesigning the architecture, fixing gradient issues, and fine-tuning reward functions—before the model finally started showing real, intelligent results. That breakthrough moment was when everything clicked: I realized just how much I love this field, and experimenting with ML officially became my main passion.
What to Expect Here
Going forward, this blog will be a mix of:
Ongoing ML and reinforcement learning experiments
Notes and thoughts on cognitive architectures
Practical takeaways, debugging stories, and engineering logs
Thanks for stopping by! If you're interested in ML, cognitive systems, or just like following along with hands-on technical experiments, feel free to connect or drop a comment.
Tags: #MachineLearning #DeepLearning #ReinforcementLearning #DevLog #Beginner
Top comments (1)
Hi mate, my name is Stanley. I'm also training an AlphaZero-style agent (Gumbel MuZero via DeepMind's mctx), trying to solve an OFC variant with 3/4 jokers—probably nobody plays it in the real world. I think I'm stuck on rollout efficiency, and my engineering skills aren't great. I'm looking for someone who could guide me through completing the solver. If you can help, I'd really appreciate it.