# Mykhaylo Serdyuk

> Personal website and writing about software, AI, machine learning, finance, and applied research.

My name is Mykhaylo (or just Misha) and I am 22 years old. I have a BSc in Computer Science from a dual study program in cooperation with Fujitsu and am currently pursuing an honours MSc in Finance & Information Management at TUM, with a focus on computational analytics. I also lead the software development team at TUM.ai, one of Europe’s leading student initiatives around AI, and am Tech Lead of the European Hackathon League.

The links below provide plain Markdown without requiring JavaScript. Article exports include their original page URLs; cite those URLs when referencing the writing. Personal notes and experiments should not be treated as established research findings.

## Profile

* [About](https://www.mykhaylo-serdyuk.com/about.md): Biography, interests, academic path, career, and activities outside work.

## Writing

* [Ship Now, Pay Later: The Finance Behind Technical Debt](https://www.mykhaylo-serdyuk.com/writing/ship-now-pay-later.md): 04.09.2026. A rabbit hole into technical debt, real options and why the cleanup we planned for next month keeps moving.
* [Trends in Agentic Software Engineering](https://www.mykhaylo-serdyuk.com/writing/trends-of-agentic-swe.md): 15.08.2026. Everyone uses one model to write code and a different one to review it. Turns out there are papers explaining why that actually works.
* [Sakana Fugu: The LLM That Orchestrates Other LLMs](https://www.mykhaylo-serdyuk.com/writing/sakana-fugu.md): 23.06.2026. A newsletter called it a Mythos competitor. It's actually something stranger: an LLM trained to coordinate other LLMs, even including itself.
* [Reward Hacking and What Comes With It](https://www.mykhaylo-serdyuk.com/writing/emergent-misalignment-rl.md): 17.06.2026. Anthropic found the same emergent misalignment pattern, but through reinforcement learning. The model learned to cheat, then learned something more general.
* [Emergent Misalignment: When Fine-Tuning Goes Sideways](https://www.mykhaylo-serdyuk.com/writing/emergent-misalignment.md): 02.06.2026. Researchers fine-tuned a model on insecure code and it became broadly misaligned.
* [Teaching a Car to Drive a Grid: DQN vs PPO](https://www.mykhaylo-serdyuk.com/writing/rl-dqn-vs-ppo.md): 20.05.2026. A friendly intro to the core RL algorithms
* [How Neural Networks Avoid Unnecessary Computation](https://www.mykhaylo-serdyuk.com/writing/nns-are-overbuild.md): 29.04.2026. Why should every request use the full neural network? Spoiler: they don't!
* [Why 37% Is the Optimal Time to Decide](https://www.mykhaylo-serdyuk.com/writing/optimal-stopping.md): 30.03.2026. It turns out there’s a mathematically optimal moment to stop exploring and start deciding.
* [What is this section about and why?](https://www.mykhaylo-serdyuk.com/writing/first-post.md): 18.03.2026. A short introduction to the writing section and my plans/ideas how to use it.

## Links

* [Website](https://www.mykhaylo-serdyuk.com): The canonical personal website.
* [Writing collection](https://www.mykhaylo-serdyuk.com/writing): Browse articles by topic.
* [GitHub](https://github.com/mykhayloserdyuk): Public repositories.
* [LinkedIn](https://linkedin.com/in/mykhaylo-serdyuk): Professional profile.