Build the artifact that beats a resume

A resume is a list of claims about work nobody can inspect. It has to be taken on faith, which is why the process around it involves so much machinery: the phone screen, the take-home, the panel, the reference call. All of it exists to verify claims.

An artifact is work someone can look at. It shortcuts the entire question of whether you can do the thing, because it is the thing. The person reading it stops deciding whether to believe you and starts assessing the work, which is a far better conversation.

One is enough, and better than five

People build portfolios of five shallow projects. One well-executed piece of work outperforms them, because depth is the signal. Five tutorials completed says you can follow instructions. One thing carried from a real problem to a working result says you can work.

There is a mechanical reason as well as a signaling one. A reviewer does not take the maximum of your five projects. They open one, form an impression inside a minute, and that impression colors the rest. A portfolio is averaged, not maxed, so a weak fourth project makes the strong first one worse.

Where this does not apply. If you are ten years in and the thing being bought is judgment across many decisions rather than the ability to build, a weekend script is the wrong altitude and will read as junior. The principle holds — show work, not claims — but the artifact changes shape. It becomes a written argument: the decision you made, the options you had, what it cost, and what you would do differently. Same rule, different object.

Pick a problem you have already had

It has to solve a problem someone actually has. Preferably one you have personally had. The Titanic dataset does not qualify, because everyone has done it and the problem is not real. A tool that removes an annoyance from your current job is worth ten Kaggle notebooks.

Take a concrete case. Every Monday you export three files from three systems and reconcile them by hand in a spreadsheet. It takes forty minutes and you get it wrong about once a month. That is a good artifact, and most people would never think of it as one, because it is small and boring and nobody asked for it.

Small and boring is the point. You understand the domain, you know the edge cases because you have been bitten by them, and you can state what it is worth in one sentence: forty minutes a week and one error a month. A reviewer can evaluate that claim. Nobody can evaluate “I built a machine learning pipeline.”

That is the opening. The rest of the lesson — the remaining sections, the exercise and the check-your-understanding questions — is behind the gate below.