Check before you trust.
Upload a screenshot. Understand the warning signs. Know what to do next.
Drag a screenshot here
or
Choose screenshotPNG, JPG or WEBP · up to 10 MB
Hide personal details before uploading.
How it works
- Step 1
Upload
Add a screenshot of a message, chat, or email you’re unsure about.
- Step 2
Review
See the warning signs, with the exact words that raised concern.
- Step 3
Act safely
Follow clear next steps to protect your money and accounts.
How Our Detection Algorithm Works
A Two-Tier Hybrid Ensemble marrying ultra-fast local Machine Learning (TF-IDF + Naive Bayes) with Gemini Multimodal Vision and real-time Mule Bank Grounding — cutting latency and API tokens by ~65% while preventing AI hallucinations.
Multimodal Ingestion & Client Pre-Processing
Screenshots, chat transcripts, or bank account numbers are validated, sanitised of EXIF GPS metadata, and tokenized.
TF-IDF Vectorizer + Multinomial Naive Bayes
Extracts character & word n-grams, calculating posterior probability vectors across 6 trained scam categories in < 1 ms with 0 API token cost.
AOC 1441 & Police Mule Account Database
Cross-checks extracted bank accounts and PromptPay numbers against 100+ verified fraudulent records. Injects ground truth priors directly into the inference loop.
Scans layout, logos, deceptive typography, and impersonation cues in screenshots.
Extracts exact suspicious phrases directly from the user's text without hallucinations.
Flags psychological manipulation: urgency deadlines, fear tactics, and fake incentives.
Why This Architecture Matters (Latency & Cost)
By pre-filtering through local TF-IDF Naive Bayes & the mule database, we cut LLM token deliberation by ~65% and anchor the model to verified evidence.
Tier 1: Fast In-App ML Model
Sub-1ms pre-classification (<0 token cost)