The model, made visible

The career sky

All 730 rated careers, drawn as one night sky. A star’s position comes from its real skill profile (35 O*NET skills + 33 knowledge areas) — careers that use similar capabilities sit near each other, and fields condense into constellations on their own, from the math, not from an editor. Brighter = higher rated; cobalt = wide AI-moat.

What is this page & how do I use it?

This is the same capability space the personal-fit score and career suggestions already use — projected from 68 dimensions down to 2 so you can see it. Nearby stars have similar skill profiles, and that’s what makes a career switch between them plausible.

How to use it: hover any star for its rating, click for the full report, and pick your interests above the map to see where you land — the marker sits at the center of the careers most similar to your profile. One caveat: only local neighborhoods are meaningful. Distances between distant constellations are an artifact of flattening 68 dimensions into 2 (details below the map).

Find yourself in the sky

Pick interests and a marker appears at the center of the careers whose skill profiles best match yours.

How much can you trust this picture?

46.3%

of each career’s 10 truest neighbors (in full 68-d space) are still its neighbors on this map

55.5% / 67%

nearest star is from the same field, here / in the full 68-d space (the ceiling)

19.3%

what PCA — the simpler method tried first — preserved. It was rejected for losing the neighborhoods (33.8%+14.7% variance explained)

Method: seeded UMAP (nNeighbors 15, minDist 0.12, seed 20260722) over z-scored capability vectors — fully reproducible from build-sky.mjs. Flattening 68 dimensions to 2 necessarily lies a little; the numbers above measure the lie. The “you are here” marker doesn’t pretend to invert the projection — it sits at the similarity-weighted center of your nearest careers, which are named under the picker. Same capability data as the fit score.