Watchlight — real-time mini-SIEM with MITRE mapping

Problem

Security dashboards are either enterprise monsters (Splunk, QRadar) or useless log viewers. For a small team, the question isn't "store every event forever" — it's "which of these events look like an attack, right now, and what do I do about it?"

What I built

A TypeScript monorepo SIEM/SOC dashboard:

The decisions that mattered

Rules in code, mapped to ATT&CK

Every detection carries a technique ID, so an analyst sees not just "brute force" but "T1110 — Brute Force" and what it implies. That's the difference between an alert and an explanation.

Sliding window in Redis

Counting events per IP over a time window is a sorted-set operation, not a table scan. Detection stays fast as ingest volume grows.

SSE over polling

Analysts watch a live board. Polling every 5 seconds means stale numbers and wasted queries; SSE means the screen moves when the attack happens.

Result

A working SOC dashboard you can read end-to-end in an afternoon — ingest, detect, stream, triage — with load testing and a Docker path included.

What it proves

I can design a detection model, think about event volume, and build an interface that a human under pressure can actually use.