Personal AI Research
The homepage shows a card and a live-app link. These are the full stories — the real prompts and schemas behind each claim, the security bugs found and fixed, and what's still rough.
Personal AI Research
The real prompt behind the multilingual extraction and credibility grading, an hourly harvester with a real cross-instance race condition it had to survive, and a Firestore security hole found and fixed the same way a sibling project's was.
Read the full writeup →Personal AI Research
How the citation classification, worker-wage-equivalent figures, and two Gemini-powered analysis modals actually work under the hood — and a real security bug this project shared with its sibling, found by recognizing the same pattern rather than waiting for an incident.
Read the full writeup →