Product · Development
Profil
A personal site instead of a résumé, for everyone: upload your résumé, then change your site just by asking.
In short
- Problem
- A one-page résumé does not say everything, it has to be redone for every job ad, and most people do not know how to build a site to complete it.
- Solution
- You upload your résumé, add what did not fit, pick a style, then change your site simply by asking. And you can paste a job ad to see what is really missing.
- What it brings
- A complete personal site without knowing how to build one, and a job-ad analysis that tells what is really missing from what you just forgot to write.
- Where it stands
- Local prototype. The journey works and the app builds without errors, but the three AI features have not yet been tested with the real API. Nothing is online.
- My role
- The idea came from my own site. I designed the journey and the rules that frame the AI: it only touches what you ask, every change can be undone, and it is not allowed to invent.
Where the idea came from
While building my own site, I realised that the one-page résumé is a format inherited from paper. It does not show a whole person, it has to be redone for every job ad, and anything that does not fit on one line disappears.
A personal site fixes all of that, provided you know how to build one. Most people do not.
The product
- Upload your résumé, then add what did not fit: a forgotten project, the detail of an assignment, what you really did in a role.
- Pick a style from a few palettes and layouts. A free choice often gives a worse result than the starting template.
- Change your site by asking: the whole site, a section or just one sentence.
- Paste a job ad to see what matches, what is really missing, and what you probably forgot to write.
The choices that matter
The AI only touches what you ask. Without that, "shorten this text" can rewrite the whole profile, and trust is gone in one go. Every change can also be undone.
The AI is not allowed to invent. No experience, number or skill that is not in the profile. And if its answer does not follow the expected format, it is rejected.
Facing a job ad, the real question is not "does it match?". It is: is this requirement really missing, or did I just forget to write it? Same finding, opposite actions. And the second case is the most common.
Where it stands
The app builds without errors and I have tested the whole journey. However, the three AI features have not yet been tested with the real API, and nothing is online. It is a prototype.
What I did
Needs definition
The starting point: a one-page résumé does not represent a person, and redoing it for every job ad is wasted time.
Product design
I designed the three-step onboarding, the dashboard with live preview and a limited choice of palettes and styles.
Rule design
Every request to the AI targets a precise area (the whole site, a section or a sentence) and can be undone.
Prompting
I wrote the instructions that forbid the AI from inventing an experience, a number or a skill.
Testing
I tested the whole journey: sign-up, publishing, unpublishing, denied access.
Assisted development
The app was built with Claude Code.
Evidence
No errors when building
verifiedThe whole app (accounts, onboarding, dashboard, job-ad analysis, public site) passes its checks and builds without errors.
- Source
- the repository's verification scripts
- Verify with
- npm run verify then npm run build, run on 30 September 2026.
The whole journey tested
verifiedSign-up, duplicate account refused, site published then unpublished, wrong password and forged session refused, protected pages.
- Source
- manual test on the development server
- Verify with
- Journey replayed request by request on the development server, on 19 August 2026.
The AI answers within a precise frame
verifiedThe AI's answers must follow the same format as the rest of the app. An answer that does not is rejected, never saved.
- Source
- repository code
- Verify with
- Read the code of the three model calls.
Limitations
- The three AI features (résumé import, requested change, job-ad analysis) have not yet been tested with the real API.
- Not online yet: the product runs locally.
- No email verification, so no password recovery.
- No PDF export yet for hiring processes that require a file.
What I take away
- Letting an AI edit a profile without limiting where it can act means accepting that a request to shorten one sentence may rewrite the whole site.
- Facing a job ad, the most common case is not the missing skill: it is the one you have but never wrote down.
What it could bring to a company
- Telling a real gap from a simple omission helps with any assessment of an application or a supplier file.
- Limiting where an AI can act and letting people undo changes applies to any tool that edits a shared document.
Technologies
- Next.js
- React
- TypeScript
- Tailwind CSS
- Zod
- SQLite
- Claude API