Technical Documentation

Interactive Multi-Layered Analysis Guide

Platform Overview

Phishing Inspector combines multiple independent detection layers operating in parallel. At the core, 3 large language models (Llama 3.1 8B, SmolLM3 3B, Gemma 2 2B) categorize emails, while external verification services, authentication validation, and behavioral analysis contribute to a five-dimensional risk assessment through a sophisticated weighted ensemble algorithm. Click on any stage in the pipeline below to see detailed information.

11
Analysis Stages
5
Risk Dimensions
15-30s
Analysis Time

Interactive Analysis Pipeline

Click any stage to view detailed information

Select a stage to view detailed information

Risk Calculation Methodology

Content Risk

35%

AI verdicts, sentiment, linguistics, keywords, impersonation

URL Risk

30%

External verification, suspicious patterns, domain reputation

Header Risk

20%

Authentication failures, routing, IP reputation

Attachment Risk

10%

File types, hashes, filename patterns

Verdict Thresholds

Phishing: ≥70
Spam: ≥60
Suspicious: 40-69
Legitimate: <40