ObsecuraWhatever I'm working on at the time, put somewhere you can actually use it.
Machine learning
Training models, building the data they learn from, and working out whether the result holds up on anything it has not already seen.
Security
Offensive security practice — enumeration, web application testing, privilege escalation and writing up findings so someone can act on them.
Software
Front ends, APIs, deployment. Whatever it takes to get something from an idea to an address you can open.
Building a detector is easy. Trusting it is the hard part.
Almost anything will separate two piles of text you already have. The difficulty is whether it still holds on text it has never seen, written by someone it was never trained on.
So the interesting part was never the accuracy figure. It was finding where the signal runs out, measuring that properly, and putting it on the page rather than hoping nobody checks.
Paste a passage and see which sentences read as machine-written, highlighted where they sit, with the reasoning shown instead of a bare number.
Paste an email and get a verdict, along with the words that pushed it either way. Trained on tens of thousands of real messages.
Runs every provenance check at once and reports each separately, keeping hard evidence apart from statistical guesswork.
Built and waiting. The signal it reads is held by the model vendor, so it stays off until a detection method is published rather than guessing at one.
Who builds this
I'm Jaffar — a cyber security undergraduate and AI/ML engineer in Sydney. I make things, and the ones worth keeping end up here.
Sometimes that is a model, sometimes an app, sometimes just whatever I wanted to understand that week. Obsecura is the shelf, not the subject.