Maj Korent, MK

Free resources

Open-source repositories anyone can take: RoboCupJunior robotics, AI coding-agent tooling, machine-learning write-ups and small utilities.

Free stuff

11 repositories you can clone, fork or ignore. No sign-up, no email gate, no course. If one of them saves you an afternoon, that is the whole point of it being here.

If you are on a RoboCupJunior team

The vision and simulation work from our national-team seasons, cleaned up and published. No datasets or competition models are included, so nobody gets an unfair run.

  • RoboCup Rescue Vision Model A TinyML victim-detection model for the OpenMV H7: letters and colored victims in real time, under 1 MB of memory, with GPIO and interrupt output for an STM32.
  • Rescue Maze Simulation (Erebus) A full Erebus simulation environment for the Rescue season, so you can test navigation and detection strategy before you commit any of it to hardware.

If you run AI coding agents

Everything I use to keep several agents working at once. All of it local, dependency-light and free.

  • Agent Session Dashboard A dependency-free local dashboard and terminal UI for Claude Code and Codex sessions: which ones are working, which finished, and which are sitting there waiting for an answer.
  • Obsidian Tracehold A structured Obsidian vault that doubles as persistent memory for coding agents: they orient themselves before they start and leave a trace of what they did, as plain Markdown you own.
  • MajBrief A morning briefing agent: it reads a vault, a task list and a calendar, pulls the tech news worth knowing, and emails one plan for the day.

If you are learning machine learning

Reproductions and write-ups, with the experiments attached, so you can disagree with the numbers rather than take my word for them.

  • Bipropagation Study An independent reproduction and component-level breakdown of bipropagation, benchmarked honestly against tuned modern backpropagation on MNIST and CIFAR-10.
  • Galaxy Classification Study A research pipeline and paper for classifying galaxies with machine learning, datasets and analysis included.
  • XOR by backpropagation The smallest honest neural network exercise there is, in TensorFlow, for when you want to see the whole thing at once.

Small tools that save an afternoon

Nothing clever. They exist because I needed them once and they work.

  • Weka Log File Analysis Turns Weka log files into learning curves you can actually read, for comparing training runs.
  • CSV Trimmer Pulls the columns you want out of a large CSV and writes a smaller one. Preprocessing, minus the notebook.
  • XRPL Bridge Monitor Watches XRP Ledger wallets: payment tracking, watchlists and enough caching to stay quick.