RedFlagged

RedFlagged scans fake police and legal threats, breaks down specific red flags, and guides safe verification through official channels before you panic or pay.

Comment

posted an update —

I’ve pushed major improvements to RedFlagged to make my detection pipeline more resilient, accurate, and accessible! Here’s what’s new:

#Algorithmic Link Deconstruction: Upgraded rules.py with pure-Python link analysis. The engine now inspects URL anatomy for raw IP destinations, suspicious/high-abuse TLDs (.top, .icu, .buzz), brand subdomain traps, and lookalike domains via Levenshtein edit distance.

Calibrated Exponential Scoring: Replaced linear point accumulation with a saturating curve ($$100 \cdot (1 - e^{-\text{raw}/38})$$). This eliminates score runaway, producing smooth 0–100 risk values that remain logically aligned with risk level classifications.

Multimodal Vision Extraction: Upgraded server-side text extraction to use Gemini Flash Vision alongside pypdf, removing fragile local OCR binary dependencies and dramatically boosting transcription accuracy on noisy or angled mobile screenshots.

WhatsApp & Family Share Export: Added a one-click formatted alert generator in the dashboard. Users can immediately copy a clear summary card to paste into family group chats or SMS to stop ongoing fraud.

Read-Aloud Voice Synthesis: Integrated browser-native Web Speech API so vulnerable or visually impaired users can have risk verdicts and safety steps read aloud.

Log in or sign up for Devpost to join the conversation.