JobShield AI

Trace recruitment claims to evidence before you trust the offer

Comment

Inspiration

Jobseekers can receive convincing recruitment messages before they have a reliable way to check the claims. JobShield AI helps them slow down, preserve what was actually said, and decide what to verify independently.

What it does

Users paste a recruitment message, preserve the source, review extracted evidence snippets, and explicitly confirm the evidence before findings are generated. Each finding uses NOTICE → MEANING → ACTION so the user can see what was observed, why it may matter, and what to verify next.

The fictional JS-001 BrightPath WhatsApp Offer demonstrates an important safety constraint: warning signs can exist while the source explicitly says no payment is required. JobShield preserves that evidence and does not invent a payment request.

The current hackathon MVP includes:

  • recruitment-message input and source preservation
  • rule-based candidate evidence extraction
  • mandatory human evidence confirmation
  • deterministic evidence-backed findings
  • NOTICE / MEANING / ACTION guidance
  • explicit uncertainty and anti-hallucination guardrails
  • downloadable case report
  • public live prototype and source repository

How it works

The submitted MVP is a browser-based HTML/CSS/JavaScript application deployed on Vercel. It uses deterministic pattern rules rather than allowing a generative model to decide whether an offer is fraudulent. Evidence snippets remain tied to the source text, and a human confirmation step is required before findings are displayed.

The workflow is:

Source → Evidence → Human confirmation → Deterministic rules → Finding → Action → Case report

This implementation is deliberately inspectable and lightweight. The repository also documents the methodology and demo flow so another developer can understand and extend it.

Why it matters

JobShield changes the question from “Does this look like a scam?” to “What exactly was claimed, what evidence do I have, and what should I independently verify before acting?”

It is designed around the Palm92 Intelligence principle:

AI investigates, humans decide.

Limitations

JobShield AI is an educational investigation aid, not a legal determination or automatic fraud-reporting service. It does not declare a recruiter, person, or job offer fraudulent. Users should independently verify employers and recruiters using trusted contact information.

Future versions can add browser OCR, independently sourced company-record checks, and model-assisted extraction while preserving the same human-review and evidence-traceability controls.

Check the claim. Preserve the evidence. You decide.

Development

Built by Faith Wright for the Global Innovation Build Challenge V2. ChatGPT assisted with planning, implementation, documentation, and testing.

Built With

Share this project:

Updates

Submission history