AI-Augmented Security Advisory Pipeline
A deployed, client-facing pipeline that turns three noisy vulnerability feeds into a prioritised monthly advisory a client can act on.
The problem
Keeping a client current on vulnerabilities was a manual analyst job every month: read the feeds, work out which CVEs actually touch the client's stack, rank them, and write it all up. It was slow, easy to fall behind on, and the interesting signal (what is being exploited right now) got lost in the volume.
What I built
An automated pipeline that does the reading and ranking, and leaves the judgement to a human at the right moment. It ingests NVD, CISA KEV, and GitHub Advisory data, vets each CVE against the client's declared tech stack, and prioritises by exploited status and severity rather than raw CVSS alone.
- Write-ups are classified into categories and rewritten in the client's house style.
- A human-reviewable JSON checkpoint sits before generation, so an analyst signs off before anything is produced.
- The output is a formatted Word report plus an Excel workbook, delivered to the client on an automated monthly schedule.
The result
It replaced a manual analyst workflow with a repeatable one, and shifted the analyst's time from reading feeds to making the call that actually needs a person. The deployment is client work, so any public version is genericised: no client names, no real stack, sample data only.