Job-Kompass · Career tech · AI-native product

A calmer way through the job hunt

Ten of ten surveyed job seekers were tracking applications with no dedicated tool, and zero of ten would trust a bare match score. The response replaces the score with a reasoned, plain-language read, and treats applying, watching, deciding later, and declining as four equally valid outcomes.

Prototype — in buildSolo — research, product, design, and build, end to end.

Role: Product Designer
Client: Self-directed
Domain: Career tech
Type: Prototype — in build

Job-Kompass case cover — a calmer way through the job hunt; prototype in build, solo research, product, design, and build — 01Cover

The problem, from research — 10 of 10 surveyed job seekers track applications with no dedicated tool, 0 of 10 would trust a bare match score, and two interviews surfaced two different jobs to be done; the paralysis sits between finding a job and deciding whether to apply — 02The problem, from research

The thesis — warmth lives in copy and pacing, not in a character; the product never pressures, treats all four decision outcomes as equally valid, and works one session at a time — 03The thesis

Discovery to directions — three live prototypes tested three ideas; the score-free reasoning read and the one-workspace direction carried forward, the guided cover-letter session is not yet built — 04Discovery → directions

The core loop — capture a job posting, read a section-by-section reasoned fit assessment with no score, and record a decision; the full loop now runs end to end — 05The core loop

Designing for an AI you can’t fully trust — a plain-language read instead of a score, a permanent non-dismissible AI-generated disclosure, and failure states specced as next, not yet built — 06Trust mechanics

Built to stay honest — design principles enforced as CI checks with tests written to fail first, every change following the same plan, spec, build, and critical-review process — 07Built to stay honest

First contact with users — two guided sessions scored against pass/fail bars set in advance; what held, a score-versus-reasoning A/B split down the middle, and a miss recorded honestly — 08First contact with users

Status and what happens next — the core loop runs end to end today, a content-complete MVP targeted for 2026-07-25, zero outcome metrics claimed by principle, and what is still to be tested — 09Status & next

Next case study — 04

From six products to one operational hub

All work