Request Radar – Classify LinkedIn invitations as recruiters or spam

Hey HN,

I built a Chrome extension that auto-labels LinkedIn connection requests based on who's sending them.

Chrome Store: https://chromewebstore.google.com/detail/Request%20Radar/pmp...

THE PROBLEM:

I receive 50+ LinkedIn requests weekly. About half are recruiters with legitimate opportunities. The other half are MLM pitches, life coaches, and "passive income" schemes.

Sorting through them manually took 20+ minutes weekly. Worse: I missed 2 recruiter messages from companies I actually wanted to work for because they were buried.

THE SOLUTION:

Request Radar scans invitation text and adds visual badges: - Green: Recruiters/talent acquisition (don't miss these) - Red: Spam (MLM, coaches, scams) - Blue: Normal professionals

TECHNICAL DETAILS:

- Vanilla JavaScript content script - MutationObserver for dynamic content - Chrome Storage API for settings persistence - Keyword-based classification (simple > complex) - Zero external API calls - entirely client-side

Performance: - 15ms average per invitation - Runs on page mutations (doesn't poll) - Lightweight: <50KB total

PRIVACY:

- No data collection whatsoever - No analytics or tracking - Settings stored locally via Chrome sync storage - Read-only - doesn't auto-accept/reject anything - Source code could be audited (considering open-sourcing)

ACCURACY:

~85% with default keywords. Users can customize to improve.

False positives: "Sales" roles sometimes flagged as spam (working on better context detection) False negatives: Generic titles like "Consultant" without other indicators

WHAT I LEARNED:

1. Chrome extensions are underrated for distribution 2. Simple keyword matching works better than I expected 3. Privacy features are a selling point 4. Users prioritize speed > accuracy (instant feedback matters)

FUTURE:

Considering: - Open-sourcing it - Analytics dashboard (local-only, shows patterns over time) - Premium tier with advanced features - Port to other platforms (Twitter DMs?)

Would love feedback on: 1. Classification accuracy improvements 2. Other platforms where this would be useful 3. Feature requests 4. Technical implementation critiques

Built this over a weekend after getting frustrated. Clearly I'm not alone - 1,000+ installs in the first week.

Happy to answer questions!