Case Study - AI Job Search for Tech Jobs You Can Actually Get

Overview

Borderless Roles is an AI job search for tech roles you can actually get. You tell it where you live and where you can work, and it reads every listing before you see it: the hiring zone, the seniority, the pay, and whether the posting offers visa sponsorship. The jobs whose rules shut you out are filtered away.

It ships as two halves that share one profile. The search is a Next.js app over curated company career boards, refreshed daily. The application layer is a Chrome extension that opens beside any posting, scores the fit, drafts the answers, and fills the form when you click. You review every word and press submit.

Search is free. Credits pay for the AI work.

The problem

A job board shows every listing and leaves the fine print to you. Two lines near the bottom of a posting decide whether an application is worth sending at all: where the employer will hire, and whether it will sponsor a work visa. Both are easy to misread. "Remote" usually means remote within one country. "Visa sponsorship available" often means the opposite once you read the paragraph under it.

The result is a long loop with a low hit rate. Filter by remote, open the posting, find the country restriction, close the tab. Repeat. When a role does fit, the application itself is another slog: the same questions, the same resume, another twenty minutes per submission.

We built Borderless Roles for our own search first. Every listing is read by AI for the details that decide eligibility, so the feed only shows roles you can apply to, and the extension takes the repetitive part of the application without taking the decision away from you.

What we did

How the search works

Listings come straight from the career pages of companies we curate around the world. That keeps spam, fake postings, and aggregator noise out of the feed. Every board is refreshed daily, and every listing is read by AI before it reaches a user, so nobody has to check the boards one by one.

The AI extracts what the fine print actually says:

• Role and seniority. Every listing is tagged with a role family and a level, so you can drop the seniority you are not applying for.
• Pay. Salary ranges are parsed into a comparable number, so a pay floor means the same thing across boards.
• Hiring zone. Remote eligibility is read from the location requirements. A role counts as remote for your country only when its hiring zone actually includes it.
• Time zone. We read the zone a team hires in, so you can keep to roles whose working day overlaps yours.
• Visa sponsorship. A role abroad is labelled as sponsoring only when the posting itself states it. It is a signal to check, not a promise, and it never comes from a government register.

Duplicates collapse. The same role cross-posted on several boards, or one employer posting it city by city, becomes a single result. When a company fills a role or pulls it, the listing leaves the results instead of lingering as a ghost job. Anything older than 30 days is archived.

Once someone signs up and sets their pay floor, the roles they want, and their deal-breakers, the feed is ranked against that profile every day. Any role can get a fit score that explains why it does or does not match.

The extension

Easy-AIpply started as its own product: a Chrome side panel that scanned a listing once and drafted many answers from shared context. We rolled it into Borderless Roles as the application layer, so the search and the application share one profile instead of two.

What carried over:

• The stack. A Plasmo + React side panel as the eyes, a Go agent with Gemini as the brain, and a Next.js portal for profile, billing, and preferences. It was chosen for cost efficiency and speed of iteration, and it still holds.
• Preference memory. When someone edits an answer or a document, we store field-answer preferences, preference text, and per-question instructions, then inject those at prompt time on later scans. No fine-tuning, just memory that nudges drafts toward the user's voice.
• Shared-context scanning. One scan reads the role, the requirements, and the questions on the form. Every draft comes out of that same scan, so nobody pastes a job description into a chat again.

What the extension does now: a fit score first, so a weak match never costs a credit; drafted answers for every question on the form; a resume tailored to the company; a cover letter built from the talking points in the listing. Standard fields fill for free, and one more click fills the drafted answers.

It runs on LinkedIn, Greenhouse, Lever, Workday, Ashby, and more, and the panel stays open when an application runs across several pages.

The boundary matters: the extension reads a page only when you click Scan, fills a form only when you click to fill, and never presses submit, uploads files, or touches captchas. It asks for the permissions needed to open the side panel and read the active tab on demand, not browsing history or other tabs.

Results

The product is live and in daily use. Search stays free, which keeps the top of the funnel open, and credits pay for the AI that costs us to run: fit scores, written answers, tailored resumes, and cover letters. A tailored resume or cover letter is one credit and includes ten rewrites for that job. Every plan is the same product, and prices are lower in many countries.

The engineering bet was to read listings with AI at ingest instead of filtering on labels. That is what makes "tech jobs you can actually get" a promise we can keep: the eligibility check happens before the job reaches the feed, not after the user has already spent an evening on it.

Try Borderless Roles

Search is free and needs no card. A free account includes a 7-day AI trial, then Basic at $9.99/mo, Pro at $19.99/mo, or a one-time 90-day pass.