Double DQN, 4 discrete actions
Descent: landing a booster
A 2D powered-descent simulator with a reinforcement-learning pilot, inspired by the Starship Flight 5 catch.
Hi, I’m Kia, short for Kiarash Karami. I study physics in Tehran, work with data, and build simulations and AI tools to see how things really work.Hi, I’m Kia, short for Kiarash Karami. I study physics in Tehran, work with data, and build simulations and AI tools.
Double DQN, 4 discrete actions
A 2D powered-descent simulator with a reinforcement-learning pilot, inspired by the Starship Flight 5 catch.
5 agents, 1 shared kernel
Five Claude Code specialists that keep a knowledge base and learn rules from corrections.
A conversational analytics agent for an online shop: ask in plain language, get KPIs and charts.

Why I build
“What I cannot create, I do not understand.”
Richard Feynman left that sentence on his Caltech blackboard. As a physics student it became my method: write the equation, compute the picture, build the thing, then explain it until the explanation is simple.Feynman’s blackboard line is my method as a physics student: equation, picture, build, explain.
More about meThe long explanations behind each project.
Why a coin flip is fully determined by its starting state, and why you still can’t call the toss.
Small physics drawings, each computed from its equation
What I’m on this month. Updated November 2026.