Remote Job Listing Eligibility and Scam Signal Checker

The problem remote job seekers face with ineligible and fake listings

A student in Lahore, Nairobi or Manila opens a remote job board and sees hundreds of roles marked "Remote". Many of them are not remote for that student at all. The real limits sit in one sentence deep in the description: "Remote (US only)", "must overlap 4 hours with CET", "we hire through our entities in the UK and Germany", "B2B contract only". Candidates spend evenings writing applications for roles they could never legally take, and recruiters reject them without saying why.

The same candidates are the main target of fake remote jobs. Scam listings promise high pay for easy work, move the conversation to a messaging app, send a cheque to "buy equipment" or ask for a training fee. Consumer protection agencies publish guidance on these patterns, but nobody reads a guidance page before replying to an exciting offer. The pain is wasted time on one side and lost money or identity documents on the other.

What already exists and the gap

• Remote OK and We Work Remotely are large remote job boards; some listings carry region tags, but many say "Anywhere" in the tag and restrict location in the text.
• Job scam checkers such as JobScamScore and LoopCV's Job Scam Detector let you paste a posting and return a risk score from a set of red-flag checks.
• The US FTC's job scam guidance explains the common patterns (upfront fees, fake cheques, chat-only interviews) in plain language.

The gap is the combination, and the honesty. Boards do not check a listing against your country, time zone and the employment types you can accept. Scam checkers give a single score, often from an opaque model, without pointing at the sentence that triggered each flag. Nobody combines "can I take this job?" with "is this job real?" in one tool that explains itself and shows its accuracy.

What you build

A web app with one main role, the candidate, and an admin view for maintaining rules:

• Candidate profile: country, time zone, the overlap hours they can give, and accepted employment types (employee via an employer of record, contractor, freelance).
• Listing input: paste the text, or import from a board's public feed where the terms allow it.
• Extraction: location restrictions, time zone rules, employment type and pay range, each with the exact source sentence highlighted.
• Eligibility verdict: eligible, maybe or not eligible, with plain-language reasons ("the listing restricts hiring to US residents; your profile says Pakistan").
• Scam signals: messaging-app-only contact, free email domains for a supposed company, fees or equipment purchases, pay far above the stated role level, urgency language. Each flag links to the triggering text and to official guidance.
• Saved listings: a small history so candidates can see which listings passed.

The technically interesting part

The hard problem is explainable extraction from messy text, measured honestly. Start with rules and patterns: phrases like "Remote (US)", "UTC-5 to UTC+1", "EST hours", "B2B", salary ranges in several currencies and formats. Normalise what you find into structured fields (a set of allowed countries or regions, an overlap window in UTC, an employment type). Then compare against the profile with clear logic, for example computing the actual overlap between the candidate's working hours and the required window in UTC.

Optionally add a language model extraction step, but always show its output next to the source sentence and run both extractors against the same labelled set. Your supervisor should see a table of precision and recall per field for the rule-based parser and the model, and a discussion of where each fails. That judgement about when to trust automated output is what makes this more than a text box and a score.

How to evaluate it

• Labelled test set: at least 100 real listings (copied manually, not scraped) plus a set of synthetic scam listings built from published scam patterns. Label each field by hand.
• Extraction accuracy: precision and recall for location, time zone, employment type and pay range.
• Verdict accuracy: compare eligibility verdicts against hand labels for 5 different candidate profiles.
• Scam flags: false positive rate on genuine listings matters as much as recall on scams. Report both.
• User test: 8 to 10 final year students each check 5 listings; record how often they agreed with the verdict and whether the reasons made sense.

Suggested plan

1. Weeks 1 to 2: collect and label the first 50 listings; write down the phrasing variety you see.
2. Weeks 3 to 5: profile, listing input, rule-based extraction with source highlighting.
3. Weeks 6 to 8: eligibility logic with time zone overlap, scam signal rules, links to guidance.
4. Weeks 9 to 11: grow the test set to 100+, measure accuracy, add the optional model comparison.
5. Weeks 12 to 15: user test, fixes, write-up with the accuracy tables.

Path to a product

The first users are your own classmates and your university's career office, which could embed the checker for graduating students. A realistic model is free for individuals with a paid version for career services and bootcamps (cohort dashboards, bulk checks, custom rules for their region). A browser extension that runs on the listing page is the growth channel. Partnerships with remote job boards that want fewer ineligible applications are a later option.

Scope warning

Do not scrape job boards against their terms. Do not declare that a listing is definitely a scam; show signals and reasons, and let the person decide. Do not give legal or immigration advice about employment status. Do not build a job board or an auto-apply bot.

Learn what you need on Sefism

• Full-stack roadmap : the order to learn front end, API and database for this app.
• Job Scams (FTC Consumer Advice) : the official patterns your scam signals should link to.
• Remote OK : a real board to study how listings describe location limits.
• We Work Remotely : a second source for building a varied test set.
• How do I find a remote software job that pays in dollars? : the user problem this tool sits inside.

Get help with this idea

The idea stands or falls on the labelled test set and the honesty of the verdicts. Tauseef can help you narrow the extracted fields, choose between rules and a model, and design an evaluation examiners will respect. If you are new here, create a free account first.

• Discuss this idea with a mentor
• Send a query to Tauseef
• Book a 1:1 session