Kubilay Yazoglu MSc.
I'm a Deep Learning Engineer specializing in LLMs, with experience in fine-tuning, optimization, deployment, RAG pipelines, and high-performance inference systems. I'm also interested in agentic AI and multi-agent systems as key directions in AI development.
Beyond LLMs, I work on Computer Vision, especially satellite imagery, and will soon share related experiments on this blog. I was active in Reinforcement Learning from 2018 to 2023, then shifted focus as Transformers took over. With RL's resurgence in LLM post-training, I'm returning to it.
If you want to know more about me like where I studied and I worked, check out the About Me page.
Contact address: kyazogluu@gmail.com
I'm planning to publish three main types of posts on this blog: AI-related content, Computer Science topics, and an "Others" section. The AI/ML section will include my own experiments, benchmarks, paper reviews, and deep dives into how specific models or methods work. The Computer Science section will focus on general software engineering topics outside the AI/ML domain. I'll mostly use it to document things I learn, kind of like a public notebook that helps me remember and explain concepts. The "Others" section is for everything unrelated to work. That includes my hobbies and personal interests like learning German, Chess (Computer Chess too!), and sports like Formula 1, football, CS2 esports, and a bit of tennis. I also enjoy gaming. While I used to be a competitive CS:GO player, these days I mostly relax with single-player games. Whatever the topic, I'll make sure the posts are worth your time and not just filler content.
Featured Posts
I made 16 LLMs play a game of deception. Here's who won (and who forgot their own name)
May 27, 2025
I tested 16 leading LLMs, such as models from GPT, Claude, DeepSeek and Llama, against each other in 100 games of social deception. Taking on roles like secret Vampires, Peasants, or the unpredictable Clown in a custom Town of Salem, they displayed a mix of clever strategy and amusing mistakes.
Pub/Sub & Queues Explained with Python, Go, and Redis
March 29, 2025
A Dockerized messaging system demonstrating Pub/Sub and Queue patterns using Redis, with a Python publisher and a Go subscriber
Analyzing Reddit Threads with Language and Vision Models
February 10, 2025
Analyze Reddit threads with LLM-powered summaries, insights, and customizable tones in a user-friendly Streamlit app.