Skip to content
AS
04 / AI & Data2026

Phishing Detector

A TF-IDF + scikit-learn pipeline that reads urgency, URL anomalies and sender spoofing at 12 ms per message, served as an API any mail gateway can call.

scikit-learnFastAPITF-IDFCI/CD
Case file
Role
Solo build — research, engineering, documentation, release
Status
v1.0.0 · 97.2% accuracy
Year
2026
Position
04 of 13
Case 04 / phishing-email-detectorArchitecture · Evidence · Signal

01 — Context

The problem worth solving.

Phishing remains the number-one initial access vector. Rule-based filters are brittle, and every false positive teaches users to ignore the next alert.

02 — Approach

How it was built.

  1. 01

    Engineer text features for urgency, impersonation and structural anomalies.

  2. 02

    Train and compare linear baselines before reaching for complexity.

  3. 03

    Serve the winning model behind FastAPI with a stable, versioned schema.

  4. 04

    Gate every change behind a CI pipeline that runs the test suite on push.

03 — Architecture

The shape of the system.

01Features

TF-IDF over subject and body plus URL and sender signals.

02Model

scikit-learn pipeline with cross-validated selection.

03Service

FastAPI endpoint returning verdict, confidence and top features.

04CI/CD

GitHub Actions runs tests and blocks regressions on push.

04 — Evidence

Receipts, not adjectives.

97.2%measured accuracy
12 msinference latency
CI/CDtested on push

05 — Hiring signal

“Security product sense plus ML engineering: it understands that a false positive has a cost and treats explainability as a feature.”

Ships working systemsDocuments the methodExplains the trade-offsKeeps evidence attached

06 — What's next

Still being built.

Add URL reputation features and a per-verdict explainability panel for analysts.