After My PhD: Lessons, Regrets, and Unfinished Research
Four years and four months after arriving in Canada, I defended my PhD. A report to my future self: what I learned, what I got wrong, and what I le...
Hadi Aghazadeh
Contemplations on AI, RL, LLMs, optimization, and scientific adventures
Welcome to my website! This is the place where I share my thoughts on AI, reinforcement learning, and any sort of scientific adventures related to these topics. I love to share my personal reflections on the books and papers I read here as well.
Applied Reinforcement Learning
Business optimization and LLM fine-tuning · Manning Publications
Most reinforcement learning material stops at games and simulations. This book is about the other part: framing a real business problem as a sequential decision problem, building an environment that respects its constraints, and getting a policy into production that people will actually trust. Warehouse logistics and dynamic pricing through to fine-tuning language models with PPO and GRPO.
In Manning Early Access Program — full release expected January 2027.
Recent Posts
Object-oriented design in the way I wish I had learned it
A tutorial on object-oriented design for ML systems: who owns which responsibility, what encapsulation is really for, and when inheritance earns it...
Reflections on Haruki Murakami running book
The one Murakami I loved. What his memoir about running says about writing, endurance and commitment — and the passages I kept.
Why I Write and What I Will Write About on This Blog
I wrote every day for years — two books, a Persian blog, a novel for learning machine learning — then stopped. Why I am starting again, and what go...
What I write about
Reinforcement learning
Framing business problems as sequential decisions — pricing, routing, allocation — and what breaks between a working policy and a deployed one.
Optimization & operations research
Mixed-integer programming, branch-and-bound and cutting planes, and where learned methods beat exact ones — or guide them.
Forecasting & ML systems
Time-series forecasting at scale, backtesting that means something, and the production plumbing that keeps a model trustworthy.
LLMs & AI engineering
Retrieval systems, agentic workflows, and fine-tuning with PPO, GRPO and verifiable rewards.
Personal reflections
The parts that aren't a method: choosing what to work on, what I got wrong and what it cost, books worth the time, and what a research and engineering career actually asks of you.