← All notesENNL

Job Scout

Finds the openings, so nobody scrolls.

  • Client — Umon
  • Status — in build
  • Scope — search, matching, tailoring
  • Team — 99six

01 — What it does

One pass through it: the search, the ranked roles, one role opened. Invented candidate, invented employers.

Job hunting is mostly scrolling. You type a job title into a board, get four hundred results, and read them one at a time to find out that most of them were never for you. The title is a bad query: it describes what you were called, not what you can do.

Job Scout takes the other end. You describe the work you want in plain words — “something with my hands, near Rotterdam, no night shifts” — and it turns that into a real search: a query plus the filters to set. Then it ranks what comes back against what you have actually done, and tells you where you fall short before you spend an evening applying.

  • Plain words in. No dropdowns to guess at, no job title you have to invent for yourself.
  • A real search out — a query and the filters behind it: remote or on-site, location and radius, hours, salary band, recency. Editable, because it is a starting point and not a verdict.
  • Roles ranked on the requirements that actually screen people out. A must-have counts three times a nice-to-have, because that is how a screening goes.
  • The gap said out loud. If a role needs something you have not done, that sentence is in the answer — before the decision to apply, not after the rejection.
  • Then tailor the CV: what to emphasise for this role, and which gap to address first.

02 — The three screens

The whole thing is three screens, and each one answers a different question. What follows is the working mock: real layout, invented data — a candidate and five employers that do not exist, because the live product runs on real people’s CVs and those are not ours to publish.

Scout — turn words into a search

Scout — what it understood, as chips you can change.

The front door. You write what you want in a box; the parsed query comes back as editable chips — role family, location and radius, hours, shift, salary band, on-site, recency — so you can see exactly what it understood and change the bit it got wrong. Beside it sits what it read out of your own history: the skills inside the work you have done, each with a count of the evidence behind it, and two roles you would never have typed into a search box.

Matches — ranked, with the reason

Matches — a fit score is worth nothing without the reason under it.

Five roles, ordered by fit, each carrying its own argument. Three matched skills, and beside every one the job you did that proves it. One line of honest gap. A tag saying whether this is a closer fit to what you already do or a step in a new direction. The filter rail keeps the count live and explains what it excluded, so a filter never quietly removes half the results.

Role — the honest read

Role — met, partly, not yet. The gap sits above the button, not under it.

One role, opened. Must-haves and nice-to-haves as two tables: the requirement, your evidence for it, and whether it is met, partly met or not yet. The gap is called out in its own block above the buttons, not buried under them. Only then the tailoring: emphasise this, address that first.

03 — How a role is ranked

A fit score is easy to fake and hard to trust, so the arithmetic is the plain kind and it is written down.

  • A must-have weighs three times a nice-to-have. Nice-to-haves break ties; they do not carry a role.
  • A deal-breaker gap — a driving licence, a mandatory certification — takes an extra penalty on top of being a missed must-have, because in a real screen it is an automatic no. The penalty is capped, so someone with several of them still has a score rather than a zero.
  • A skill is not one number. It is scored on six dimensions against written anchors, so what a given number means is fixed in advance and an unusual route through a career is not punished for being unusual.
  • Four of those six can be read off a CV — impact, breadth, experience, recency. Two cannot: how deeply someone understands a thing, and how independently they use it. Those stay unscored until a conversation establishes them, and unscored is an honest state, not a hole.

We don’t know is the absence of a measurement, never a low number standing in for one.

The rules are in the system rather than in the tone of voice. It never invents experience to fit a role: a weak point is handled by emphasis and framing — the same facts told better — and never by fabrication. It builds a search only from what you said and what is on the screen, so there are no invented companies and no invented result counts.

04 — Why it looks like this

The first version of this idea was a dark, scrolling story: beautiful, and the wrong room for it. Somebody comparing five roles at eleven at night is doing admin, not reading an essay, and a page that animates while they work is in the way.

So it was redrawn as a tool. White ground, cool greys, deep navy for anything you can press, and exactly one accent — a teal reserved for fit scores and the word “met”, so a colour always means the same thing. Amber is status and only status. Numbers are set in tabular figures so a column of fit scores lines up. Cards carry a one-pixel border and a soft radius instead of a shadow, because a screen with twelve cards on it and twelve shadows under them reads as noise.

It holds at 390 as well as at 1440: the filter rail becomes wrapping chips, the evidence column drops out of the requirement tables, and the buttons go full width. Nothing is hidden behind a menu that a thumb cannot find.

05 — Where it stands

In build. The scoring model, the ranking rules and the four conversations behind them are real and running inside Umon; what you see above is the corporate rework of the interface, captured from a self-contained mock. The candidate, the employers and every number in it are invented. The design is the deliverable here, not the data.

If that is your week too, thirty minutes is enough to find out.

Book 30 minutes →