CareerCompass — a career answer you're allowed to argue with

A man sitting alone on a sofa late at night, phone face down on his knee, thinking

“Should the next ten years look like the last ten?”

The question every career tool answers with a horoscope

The problem

Career advice is either a horoscope or a haystack.

Both feel like progress. Neither closes the gap between what you've done and what you should do next.

Personality tests

They stop at the adjective

You get four letters and a flattering paragraph. Nothing in it has read your resume, so it can't tell a 22-year-old floor lead from a 44-year-old project director with the same letters. Same test, same answer, two completely different next moves.

Job boards

They assume you already decided

Search needs a job title. But “what should I search for?” is the question — and if you knew the title, you wouldn't need help. So people search the title they already have, and get offered the career they're trying to leave.

The AI ones

They won't show their work

A confident paragraph appears. Why that job? Why 91%? What if you'd left one thing off your resume? No answer — and a recommendation you can't interrogate is a recommendation you can't trust with the next ten years.

The solution

Two signals, combined — and shown.

CareerCompass scores every path against who you are and what you've already proven, then hands you the receipt.

Traits

Your wiring, as a shape — not a label

Five sliders become an interest profile across six dimensions. We compare the shape of your profile to the shape of each career, so someone moderate on everything still gets a real answer instead of “you match all 18 equally.”

Evidence

Your resume, as proof of work

Paste it or attach it, and the engine extracts what you can demonstrably do. Add industry research, job postings and notes as you dig — every document you add sharpens the scores and feeds the coach.

Traceability

Every number opens up

Each path shows its interest profile, your fit percentage, your skill overlap, and exactly which of its core skills you already bring. Disagree with a score? You can see precisely what produced it.

Compass

A coach that has read your file

Compass talks — out loud, if you want. It's grounded in your own profile and documents, so it answers about your situation. It is also instructed to refuse to invent numbers it doesn't have.

Built for the person mid-decision

Not for HR, not for recruiters — for the Sunday-night version of you wondering whether the next ten years should look like the last ten. Four ways in, whichever is least effort: tap through eight questions, attach a file, paste text, or just talk — the app transcribes.

How it works

Four minutes to your first ranking.

No onboarding call, no consultant, no waiting on a report. You can be arguing with your own numbers before the kettle boils.

A thumb tapping a large answer card on a phone screen

Step one

Answer eight tappable questions

Plain questions with big answers — no jargon, no sliders you've never heard of. They derive your five trait values for you, and you can still edit every one by hand afterwards. Or skip the wizard entirely if you already know your type.

Overhead view of a printed resume beside an open laptop, one hand on each

Step two

Hand over the evidence

Paste your resume or attach a file, and watch the extracted skills appear as chips you can check. Keep adding — industry research, job postings, your own notes. Twelve documents, and every one of them moves the scores.

A woman pointing at one highlighted row in a ranked list on her laptop

Step three

Read the ranking — and the arithmetic under it

18 paths, scored and ordered, each one carrying its fit percentage, its skill overlap, its AI-proof score, a BLS median wage with the year attached, and an honest note on whether the credential is an exam or three years of school.

A man walking a hallway at night wearing earbuds, talking to the voice coach on his phone

Step four

Talk it through, then track it

Ask Compass why a path ranked where it did — out loud, hands free, while you walk. Then log the applications, the follow-ups and what came back, so the next decision starts from a record instead of a memory.

AI-generated — not real users

See it work — no sign-up

Three people. Three industries. Three stages.

These rankings are computed in your browser right now, by the same engine you get — not screenshots. Open any one to explore the full profile.

The actual product

Rendered live. Never screenshotted.

Both panels below are drawn right now by the same engine and the same bundled dataset the app ships — a screenshot can go stale; this cannot disagree with the product.

Career Paths — top card for the Devon example

Computing…

Explore any field — live search: “solar”

Computing…

The reason this exists now

“Am I about to be replaced?” deserves better than a vibe.

Here is what the primary sources actually say — including the parts that cut against the panic. Every figure below is linked; nothing here is our estimate.

49,135

US job cuts citing AI in 2026 so far — and in April AI was the single leading reason given for layoffs for the second month running (21,490 cuts, 26% of that month's total). Challenger, Gray & Christmas, 7 May 2026

54,836

Cuts that cited AI in all of 2025 — about 5% of the year's total. 2026 passed most of that in four months, and AI's share of announced cuts rose from 13% through March to ~16% through April. Same source

↑ wage

Exposure now rises with pay, not the other way round. Capability-based measures put cognitive, analytical, administrative and managerial work at the top — the reverse of the old “routine manual work goes first” assumption. ILO Research Brief, Feb 2026

The part nobody puts in a headline

The same ILO brief is blunt that exposure indices measure what AI could technically do — not whether firms will adopt it, whether it's profitable, or what happens to employment. In its own words, they “cannot be interpreted as predictions of job displacement.” Anyone selling you a percentage chance your job dies is selling you something.

It also finds a second-order risk that task-level scores miss entirely: exposed analytical and administrative roles sit in dense hubs of the occupational network, so displacement spills into adjacent jobs — auditing, compliance, office administration, project coordination — even when those jobs don't look automatable themselves.

So what do you actually do with that?

Two things follow, and both are things software can help with:

1. Know your direction of travel. Every path here carries an AI-proof score built from the factors the literature agrees matter — and when you plan a move, we show whether it takes you toward safer ground or away from it. Sometimes the exciting job is the exposed one. We say so.

2. Know what transfers. Displacement panic assumes you start from zero. You don't. The engine counts how much of a target role you already evidence — usually most of it — and names only the gap.

The AI-proof score

Five factors, published weights, no crystal ball.

We don't have a licensed occupation-level exposure index, so we don't print one and pretend we do. Instead each path is rated on five factors the research agrees on, and you see the factors — so you can argue with the inputs, not just the answer.

What raises and lowers the score

Factor	Effect
Routine — predictable, repeatable text and data work	↑ exposure (40%)
Network centrality — sits beside lots of exposed work	↑ exposure (20%)
Embodied — needs a physical presence	↓ exposure
Interpersonal — trust, care, persuasion with real people	↓ exposure
Accountable — a named human must answer for the decision	↓ exposure

The three protective factors are averaged and carry the remaining 40%. Scores across the 18 paths currently run 13 to 76.

Where the factors come from: the direction of findings in the ILO's February 2026 review of AI exposure measures — including its network finding, which is where “sits beside a lot of exposed work” comes from — read alongside announced-layoff reasons from Challenger, Gray & Christmas. The ratings themselves are ours, and we say so.

What it is not

It is not a probability that your job disappears, and it is not a sourced index. It is our reading of the direction the published research points, made explicit so you can check it. If you think a role's “interpersonal” rating is wrong, you can see that it's rated 2 and disagree — which is more than a confident paragraph from a chatbot lets you do.

The transition model predicts effort — not pay

Pick the role closest to what you do now and how many hours a week you can study, and every path tells you: how much of it you already have, exactly which skills are missing, roughly how many weeks that is at your pace, whether the credential is exam-only or real school, and whether the move makes you more or less AI-exposed than today.

It will not project your future salary. A BLS median is an anchor for a whole occupation, not an offer for you — and the invented number is precisely the one you'd bet a career on.

Why the algorithm is worth trusting

It's arithmetic you can check — not a black box.

The whole scoring model fits in a paragraph. That's on purpose: a career decision you can't audit is one you'll quietly ignore.

The score, in full

70% — trait fit

30% — skills

Every path score is 0.7 × trait fit + 0.3 × skill overlap.

Trait fit is a centered cosine similarity: we subtract each profile's own average before comparing, so what gets matched is the shape of your interests, not their volume. This matters more than it sounds — before we centered it, every path scored 82–96% and the ranking was noise. Now the spread across 18 paths is 48–67 points.

Skill overlap is the share of a path's core skills your documents actually evidence. Every path lists a comparable number of skills (7–9), so no career can win just by being vaguely described.

What most tools do	What we do
Ranks you on personality alone	Combines traits with resume evidence
Hides the model	Publishes the weights on the page
Everything scores ~90%	Real spread, so the ranking means something
Quotes salaries from nowhere	Every wage is BLS-sourced, year attached
Generic AI chat	A coach grounded in your own documents
Your resume becomes their product	Your record is yours, readable only by you

The honesty test we hold ourselves to

The founder ran his own real resume and real 16-personality result through it as Subject 0 — published inside the app, contact details stripped. The engine's top answer for him is UX Researcher — a role he has never held and doesn't sell — with Software Developer third. We publish whatever the engine computes, and when we swapped our hand-estimated interest profiles for O*NET's published ones in v1.4, his ranking moved and this sentence moved with it — the page follows the engine, never the other way around.

Social proof

We have no testimonials. That's the point.

What we could have put here

Three smiling stock portraits, three invented names, three quotes about how CareerCompass changed someone's life. It takes about ten minutes and it is what most pre-launch products do.

We're not doing it. This product's entire argument is that you should be able to check our numbers. A page that opens with a fabricated human being has already lost that argument — and in the US, inventing consumer testimonials isn't just tacky, it's illegal.

What we have instead

A worked example on a real person. The founder's actual resume and actual personality result run through the engine inside the app — including the result that doesn't flatter him.

Three example profiles you can run yourself. Not screenshots: the model executes in your browser, and you can open any of them and change the inputs.

The model, published. Weights, factors and limitations are on this page. You don't have to take a stranger's word for it, which is what a testimonial is.

Two people sitting side by side at a table, discussing career results on one laptop

Want to be one of the first real ones?

Founding members who tell us honestly what worked are how this section eventually fills — with attributed quotes from people who exist

Try it and tell us

AI-generated — not real users

Founding window closes —

Free — for life

Everything is free while we prove the concept. Create your account before the date above and your account stays free permanently — including after paid plans launch.

Full 18-path ranking with every number shown
All 923 O*NET occupations, searchable, with BLS pay and outlook
Resume plus a 12-document library
Compass, the AI voice coach
Application tracker, map and network tabs
Free forever, grandfathered — not a trial that expires

Claim founding access

We're not naming a future price yet, because we haven't set one and we won't invent a number to create urgency. The date above is real; the grandfathering is recorded on your account the moment you join.

Straight answers

Questions worth asking before you upload a resume.

Who can read my resume?

You. Your profile is stored under your own account and the server rules allow only the owner of a record to read it — not other users, not our team. Before you sign in, everything stays in your own browser. If you use the AI coach, a clamped excerpt of your profile goes to Google's Gemini and the reply text goes to ElevenLabs to be spoken; the app tells you that on the coach screen, and the coach is optional.

Is this a real psychometric test?

No, and we say so in the app. The five sliders are a preference indicator, not a validated instrument, and the mapping from traits to interests is a documented heuristic. It's one signal, weighted at 70%, next to hard evidence from your resume at 30%. What IS published data: every career's interest profile now comes from the O*NET 30.3 database (923 occupations, CC BY 4.0) — the profiles are real even where your side of the comparison is self-reported.

Where do the salary figures come from?

Every wage in the app is the U.S. Bureau of Labor Statistics median annual wage (May 2024, national), and every outlook figure is the BLS 2024–34 employment projection — bundled from Employment Projections Table 1.2 with the year printed next to the number. Where BLS publishes no figure for an occupation, we show none. What we will not do is predict your salary: a median is an anchor, not an offer.

Are the people in the photographs real?

No. Every photograph on this page is AI-generated, commissioned for CareerCompass, and labelled as such wherever it appears. Nobody in them is a real user, a real customer or a real employee, and none of them is offering you a testimonial — they illustrate the moment the product is for, nothing more. The portraits on the three example cards are different: those are licensed stock photographs of real people who are unconnected to this product, and they are credited by name.

Are the three example people real?

No — Maya, Devon and Priya are illustrative composites built to show the engine across three industries and three career stages. That's labelled on every screen they appear on. The one real person in the app is the founder, published as Subject 0 with his consent and his contact details removed.

What happens when the free period ends?

If you joined before the founding date, nothing — your account is marked as founding when you create it and stays free. For people who join later, we'll announce pricing before we charge anyone, and nobody gets billed without opting in.

Can I delete everything?

Yes. Signing out clears the local copy, and your cloud record is deletable by you — the permission to delete your own profile is written into the server rules, not a support request.

Your next move is already in your resume.

You just haven't seen it scored against everything else yet.

Map my career — free