spoorthinavale4-cmyk/phishguard-security-copilot
AI-powered phishing detection copilot, multi-layer URL analysis with ML, Google Safe Browsing & LLM-generated SOC explanations
What's novel
AI-powered phishing detection copilot, multi-layer URL analysis with ML, Google Safe Browsing & LLM-generated SOC explanations
Code Analysis
13 files read · 4 roundsExtracts URLs from pasted email text, runs them through a rule-based + Random Forest ML pipeline with Google Safe Browsing override, and returns risk verdicts with LLM-generated SOC-style explanations via a FastAPI backend and React frontend.
Strengths
The multi-layer detection approach (trusted domains → brand impersonation → risky TLDs → ML model → Google Safe Browsing) is a reasonable defense-in-depth architecture, and the feature extractor performs real network I/O (SSL socket, WHOIS, DNS) rather than pure string matching. The training script is well-structured with stratified splits, class weighting, cross-validation, and feature importance reporting.
Weaknesses
12 of the 30 ML features are hardcoded to 0 at inference time (Favicon, Links_in_tags, SFH, Redirect, Iframe, etc.), creating a significant train/serve skew that undermines the model's reliability. There are zero tests, a hardcoded API key in check_models.py, broken dead-code files (pipeline_demo.py, risk_explainer.py), debug prints in production code, and duplicated domain lists across modules.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
60
Contributors
1
Files
43
Active weeks
3
Repository
Language
Python
Stars
1
Forks
0
License
—