Building Aura: how matching learns what you value

Most job matching is keyword matching wearing a nicer interface. You say “content strategy,” the system finds postings that say “content strategy.” That works right up until it doesn't, which is most of the time, because a job title tells you almost nothing about whether you'd want the job.

Start with values, not keywords

The questionnaire is built to capture what a job title can’t: how much autonomy you want, whether you do your best work in structure or in ambiguity, what kind of impact makes the work feel worth doing.

Those are the signals that actually predict fit. They’re also exactly what a résumé is built to leave out.

Turning answers into signal

Your answers resolve into a structured profile, and every match resolves to a 0–100 score with a plain-language rationale attached. The score has two axes, and we show both:

• Competency: can you do the day-to-day work?
• Values fit: will you thrive in this environment, given what you told us matters?

We keep those legible on purpose. The score isn’t a black box: you can see which axis is pulling it up or down.

What we haven’t solved yet

A sparse profile produces a cautious score. With thin input the model hedges, and it should. We’d rather wait than guess. The model is instructed never to assume a skill or value you haven’t told us about.

Why legibility matters

A score you can't interpret is a score you can't trust. We'd rather show the work, including where it's uncertain, than hand you a number and ask you to take it on faith. The model gets better with more data. The principle doesn't move: you should always be able to see why.