Lion Blau
IN ACTIVE DEVELOPMENT
Job searching, redesigned as a system
Arco scores every role against your profile and coaches you from first match to offer — honest fit scoring, interview prep, tailored pitch, no noise.
This case study documents a system mid-build, not a finished ship.
Role: Product, design, and AI-assisted build · 2026 — ongoing
2026 — ongoing
Role: Product, design, and AI-assisted build
Honest scores over flattering ones
A fit score only works if users can trust a low one.
Coach, don't generate
Tailoring that starts from your real experience, not a template.
The pipeline is the product
From first match to offer in one place — feed, prep, and tracking connected.
Built with the tools it runs on
Designed and shipped solo with Claude Code and Figma MCP.
Every role opens as a briefing, not a listing: what the company actually does, what it pays, and why the score is what it is.
Context
Job searching is a full-time job nobody designed
Arco started inside my own job search. Looking for a senior design role, I kept running into the same problem: the Israeli tech landscape is full of companies I'd never heard of, many doing genuinely innovative work — and no tool existed to quickly understand what a company actually does, whether the work is interesting, and whether I'd be a real fit. Job boards give you volume; they don't give you comprehension.
So I built the tool I needed: a feed that scores every role against your real profile, explains the fit instead of just asserting it, and an AI coach that takes you from "who is this company?" to a tailored application and interview prep. It's live at arco.careers — and this page is deliberately a snapshot of a product in motion.
Mobile came first, under an earlier name. Desktop overtook it — and the phone now has to be rebuilt to match the product it became.
CHALLENGES
The hard parts are still hard
Honest scoring is harder than flattering scoring. The easy version of a fit score tells everyone they're a great match — it feels good and means nothing. The version worth building has to tell you a role is wrong for you, without the product feeling discouraging to use. Calibrating that honesty is ongoing work, not a solved problem.
Job sourcing is fragile by nature. Boards block crawlers, change structure without notice, or render in ways that break parsing entirely. Two sources have already been decommissioned; keeping a clean, deduplicated, correctly classified feed alive is continuous maintenance, not a one-time integration.
Every score costs money. AI scoring at feed scale means real API economics. The system pre-ranks candidates before the model sees them, and routes everyday tasks to a lighter model while reserving the stronger one for judgment-heavy work like CV merging. Where those seams sit is a live design decision, revisited constantly.
One designer, two form factors. Arco started mobile-first. The desktop web app then overtook it — it's now the stronger, more complete surface — which means the mobile experience has to be rebuilt to match what the product became. That's the honest cost of learning in public: the first surface you build is rarely the one you keep.
A score that flatters everyone is a score nobody checks twice. Fewer, harder matches are what make the number worth reading.
The coach runs on its own dark surface, scoped to read as a different kind of space.
Each conversation is bound to one role, and tailoring is tracked per job — 78 → 85 — without ever touching the master CV.
WHERE THIS STANDS NOW
Built, open, next
Working today:
Arco is live. The scored job feed, the AI coach (CV analysis, per-job tailoring, guided optimization), and the pipeline board all function end-to-end. The desktop web app is the primary surface, with free and Pro tiers defined.
Still open:
The mobile app is being rebuilt to catch up with desktop. Payments are in activation review with merchant-of-record providers, so subscriptions aren't switched on yet. Job sources are being expanded lane by lane.
Next:
Payments live, mobile redone, sourcing broadened — in that order.
AI-NATIVE BY DESIGN
Designed with AI, for AI
Arco doubles as a working demonstration of how I design now. There is no engineering team: every feature is specified as a structured prompt and executed through Claude Code, with Figma MCP bridging design intent and implementation. Design decisions — from color semantics to which model handles which task — are written, versioned, and shipped by one person.
That includes designing for AI, not just with it: model selection, cost seams, prompt architecture, and the UX of machine judgment (how a score earns trust, how a coach corrects without condescending) are product design decisions here, as much as any screen.
I design systems, not just screens—products that stay clear under real-world use.
