Evidence, not adjectives.

Making an AI workflow auditable: 15 tasks, 15 honest classifications

• ProblemNo matter how advanced AI gets, it can't mimic the authentic feel and point of view of a human being. Yet most people treat AI like magic copypasta - no thought about which tasks actually deserve it. I wanted to see where my time really goes, where I could squeeze in new tasks, and what I should cut. That takes an audit, not a vibe.
• ApproachI listed 15 real weekly tasks and classified each honestly: Just me / Delegate with review / Collaborative / Fully automate. Six stayed "just me" because the experience matters more than the result. For measurable work - lectures, quizzes, assignments - AI handles repetitive extraction and I keep judgment. My model: offload the repetitive, keep the creative. Then I wrote "done well" definitions before doing the work, in numbers.
• OutcomeA workflow someone else can audit and check, not just listen to - the same discipline I'd bring to a team lead who needs decision-ready reports instead of hand-waving. Standards: notes reviewed 90% before use, quiz scores at 80%+, assignments submitted 24 hours early with every claim checked against two or more sources.

Artifact: full workflow-audit captures being added during launch week - the source document exists and every number above traces to it. Nothing is faked.

Binary Triage - the analysis approach, running

This site carries one real feature instead of a screenshot carousel: paste raw strings or hash output from an unknown file, and Binary Triage extracts indicators - hashes, IPs, URLs, registry persistence keys, mutexes, encoded blobs - and drafts a structured report skeleton with behavior hypotheses and recommended next steps. It runs entirely in your browser; nothing is uploaded anywhere.

Open Binary Triage →

Building a proof statement that survives the first five seconds

• ProblemA résumé line is just a piece of information. Millions write "reverse engineering" on a CV. A claim holds no value if nothing practical sits behind it. I needed something a stranger could check, not just read.
• ApproachMy first draft was "cyber security student who uses AI to accelerate reverse engineering" - three skills hiding behind each other. Narrowing it reflected a real decision: a specific skill shows inner confidence; a vague one doesn't give the audience a clear image. So it became one skill, one outcome, no "and": reverse-engineer malware and binaries into clear, actionable analysis. Then I named who matters - a malware analysis team lead at a security vendor - and the one action that works: book a short call to review my approach.
• OutcomeA pressure-test of my four-page sitemap caught something my first draft overlooked: the conversion moment was buried. The visitor convinced on the Work page had no exit. The fix - a second call-to-action right under the cases, where conviction actually happens - is a decision you can verify by scrolling this very page.

A real binary walkthrough: unknown sample to recommendations

• PlanOne real sample taken end-to-end: static triage, disassembly of the interesting logic, findings, and recommendations - documented with genuine Ghidra and terminal captures, published with its repository.
• StatusAnalysis is underway. Rather than fake it, this slot stays honest until the work exists: write-up coming. When it ships, it takes the lead slot from Case A - per the plan set in Week 3.