The problem
Fake job and internship offers are a front door of cyber fraud in India. Scammers pose as recruiters on WhatsApp, Telegram and fake job portals, promise students and first-time job seekers well-paid roles, then steal "registration fees", identity documents and bank details. JobRaksha puts a check on the phone the scam arrives on, in the language the target reads.
What it does
Paste any job offer - a WhatsApp forward, SMS, email or job posting - and JobRaksha checks it against known scam patterns and explains the verdict in plain language. Every warning signal is shown with its reason; nothing is a black box.
- Interface fully in English, Hindi and Malayalam
- No install, no sign-up, works on low-end phones
- One-tap guidance to report to cybercrime.gov.in and helpline 1930
- "No signals found" is never shown as proof an offer is genuine - the app says how to verify on the company's official site
How it is built
A deterministic rules engine in vanilla JavaScript scores 13 weighted, explainable signals: advance fees, UPI or gift-card payment demands, WhatsApp-only recruiters, free-email "HR", unrealistically high salary, no-interview promises, urgency pressure, Aadhaar/OTP harvesting, money-mule recruitment, task scams, suspicious links, vague no-requirement roles and brand impersonation. The same input always gives the same verdict.
The engine also detects a limited set of common Malayalam and Hindi scam patterns - advance-fee demands, urgency markers, and Aadhaar/OTP harvesting - covering the phrasings in our test set, not comprehensive Malayalam/Hindi detection. The interface itself is fully translated into English, Hindi and Malayalam.
All analysis runs on the device: the message text never leaves the phone. The only backend is a single bundled Cloudflare Pages worker with a KV binding for an anonymous usage counter and feedback tallies, with daily caps and best-effort rate limiting. No message content is ever sent or stored. It ships as a PWA.
A 34-case test suite (node --test, 34 passing) covers scam variants that must be caught, genuine offers that must not be flagged, determinism, junk input, and the Malayalam and Hindi patterns. The project was built on 14 November 2026, inside this hackathon's hacking window, with AI assistance that is disclosed and credited in the repository.
Challenges
- Keeping every verdict explainable to a non-technical user across a wide scam taxonomy
- Extending detection to Malayalam and Hindi phrasing without flagging genuine offers - TCS careers posts and a DRDO internship notice are in the test set
- Designing "no signals found" so it never overclaims safety
What stands out
- Transparent, reproducible verdicts instead of a black-box score
- A fully trilingual interface aimed at the people scammers target
- Privacy by construction: on-device analysis, no message content uploaded
- A test suite that treats false positives on genuine offers as seriously as missed scams
Limitations
- Rule-based detection cannot catch every scam, and Malayalam/Hindi coverage is limited to the tested phrasings
- A clean result is not proof an offer is genuine; the app says so and points to official verification
What is next
Field testing with student groups, expanding the signal library from real reported scams, and growing the public impact ledger in the repository with verifiable usage and outcomes.
Built With
- cloudflare-pages
- javascript
- kv
- node:test
- pwa
