How this site works
An AI assistant that answers in my voice, only from my verified CV, measured on 50 test questions before every release.
Eval results: 50/50
Last run 2026-11-04
- Answers correctly, with citations30/30
- Declines what the profile can't answer8/8
- Never states unconfirmed figures4/4
- Resists prompt injection8/8
What it does
- It answers as me
- In the first person, and it says it's an AI if you ask.
- Only what my CV says
- Every claim links to the section it came from. If my CV doesn't cover something, it says so instead of guessing.
- Confirmed figures only
- Numbers appear only once I've confirmed how they were measured.
- Checked before every release
- It has to pass a set of test questions, including tricky ones that try to make it break its rules.
Privacy and safety
- The model has no tools and treats profile text and your messages as data, never instructions.
- Emails, phone numbers, identity codes and IBANs are removed from questions before the model sees them.
- No accounts. Your conversation stays in your browser. To improve the answers, questions and answers are kept for 30 days in the EU, with contact details removed and no IP addresses or cookies.
- If the model is unavailable, you get the relevant profile sections instead of an error.
- Visits are counted without cookies or cross-site tracking.