RedFlag Dojo

Practice mode: three fictional recruiter messages. Tap the suspicious clauses, submit, and earn a belt based on your score.
Analyze results: highlighted phrases match numbered evidence entries, each with its tactic, why it matters, and what to verify.

Inspiration

Many freshers in India get first-contact recruiter messages on WhatsApp, Telegram, LinkedIn and email. Some of them are fake, asking for a registration fee, pushing the chat off the platform, or requesting Aadhaar and bank details too early. Most scam checkers just say "scam" or "not scam", so the user learns nothing. I wanted a tool that teaches people to spot the signs themselves.

What it does

RedFlag Dojo has two modes.

Analyze: You paste a recruiter message. The app highlights each phrase that matches a rule, names the tactic, explains why it matters, and says what to verify. It then gives a signal level (few, some or many) and a checklist built only from the signals that fired.

Practice: You get three fictional messages, two suspicious and one clean. You tap the clauses that look suspicious, submit, and see what you caught, missed, or flagged by mistake. Your score moves you through White, Yellow, Brown and Black belts.

The app never says a message is a scam or safe. It shows signals, because it cannot verify whether a company is real.

How we built it

React and Vite, in plain JavaScript, with no backend, database, accounts or paid APIs
Rules are written in a JSON file: 12 rules across 6 categories. A small, transparent rules engine finds the matches.
Scoring and belt logic are pure functions, so they are easy to test.
107 automated tests, plus a data check that every fictional scenario follows its rules.
Deployed as a static site on Netlify.
Planned first with the Devpost Learn Skill Pack: scope, PRD and technical spec, all in the repository's devpost folder.

Challenges we ran into

Keeping the rules honest. Phrases like "urgent" or "good pay" can appear in legitimate messages, so I tightened the phrase lists and added clean messages to the tests.
Scoring fairly. Flagging everything had to score badly, and a clean message left untouched had to score full marks. I changed the penalty for false flags to make the scoring match the design.
Scope. I cut features like company verification and formal offer letters, because they needed outside data or would not finish in time.

Accomplishments that we're proud of

A working app, deployed and tested on desktop and phone.
Practice mode that teaches restraint as well as detection.
Every flag shows its evidence, so nothing is a black box.

What we learned

Planning before coding made the build faster and the scope easier to defend. Transparent rules are easier to test and explain than a model, and they fit the honesty goal. Writing fictional examples that are clearly fake, but still realistic, took more care than I expected.