PROJECTS · SIMPLE TO ADVANCED

Build things people actually use.

Each project turns course lessons into a working artifact: a report generator, a memory server, a task queue, an agent. You build it stage by stage, a grader checks every stage, and a tutor walks you through it.

Start the first project Submit your project
HOW IT WORKS

Three commands, one project

Choose Python, Rust, TypeScript, or Go for the job. Core tests run offline; optional framework labs list their dependencies.

  1. Create your workspace

    Copy the starter files. Each stub gives you the exact signature to implement.

    python3 scripts/project_test.py research-report-agent --init my-report-agent

  2. Run the grader

    Build one stage at a time. The grader checks your work and every earlier stage.

    python3 scripts/project_test.py research-report-agent --stage 1 --path my-report-agent

  3. Learn with the tutor

    Want guidance? In Claude Code, work through each stage with hints and feedback.

    /build-project research-report-agent
THE LADDER

Pick your level

Five levels. Each adds one new kind of engineering. New projects join a level as they are written or submitted.

COMMUNITY

Built something useful? Add it to the ladder.

Community projects use the same layout as core ones: a project.json, stages with a lesson, a starter, and tests, plus a reference solution. They appear here with your GitHub handle.

  • Produces a real artifact, not a chat demo
  • Every stage has tests that pass offline, with no API key
  • Independent evaluation fixtures and clear limits on what the score proves
  • Original implementations with links to official specs and research

After copying, set id in projects/my-project/project.json to my-project before running the grader.

cp -r projects/_template projects/my-project
python3 scripts/project_test.py my-project \
  --all --solution --strict
Read the submission guide