Build a project that shows your skills

Create a small, reviewable project and document what you actually contributed, including AI assistance.

Choose a small problem with a real user

Describe who the project helps and the task they need to complete. Define a narrow version that works from start to finish. A small, tested project is easier to review than a large feature list with incomplete flows.

Keep the evidence reviewable

Write a README with setup steps, a working example, known limitations and the decisions you made. Add tests for important behavior and show what happens when inputs are missing or invalid. Keep secrets and personal data out of the repository.

Link to a repository you control or are allowed to share. Explain which parts you built and which parts came from a tutorial, library, teammate or assistant. A link alone does not prove authorship.

Use an assistant thoughtfully

You can use an assistant to discuss architecture, inspect errors or practice explanations. Keep responsibility for understanding and checking the result. Record substantial assistance so a reviewer can understand how the project was made.

Sonder's scoped integration can save project metadata through a compatible MCP client. This does not observe your editor, execute your code, prove authorship or verify your skill. Project evidence remains self-reported until a separate review is performed.

Prepare to explain the project

Be ready to walk through a failure case, a tradeoff and a change you would make next. The ability to explain and modify your work is useful evidence beyond a polished screenshot.