Descent: landing a booster
A deterministic 2D powered-descent simulator with a Double-DQN pilot, inspired by the Starship Flight 5 booster catch.
Nine things I built to understand something: physics simulations, AI agents, MCP servers and data work. One featured below, eight in the grid.
Featured
A coin is not random: one exact starting state gives one exact landing. This lab solves the rigid-body motion from release to first contact, then shows how a tiny measurement error can still cross the line between heads and tails.
A deterministic 2D powered-descent simulator with a Double-DQN pilot, inspired by the Starship Flight 5 booster catch.
Five Claude Code specialists (design, engineering, research, business, usage) that learn rules from corrections.
Ride quotes, booking, live driver tracking and two-stop rides for Iran’s Snapp, from inside an AI assistant.
Search freelance projects, bid and reply to clients on Ponisha from Claude or Cursor. Bilingual, Persian first.
A local MCP server and Figma plugin that let Claude read and edit the open file. This site was designed through it.
Bootcamp capstone: an RFM win-back strategy whose A/B test on simulated data lifted reactivation from 7% to 17%.
Customer segmentation and an experiment that raised GMV 10%; a funnel fix that cut login drop‑off by 10%.
A conversational analytics agent for an online shop. Ask in plain language; it picks from approved metrics over a 190-table WooCommerce database and live site events, and answers with KPIs, charts or tables. Self-hosted and private.
Open source
The code for the public projects is on GitHub. Issues and ideas are welcome.
github.com/Kia-Karami