2026 – ongoing · Remote
bernaz.fr — a trilingual, eco-designed and accessible career site
Entrepreneuriat personnel — Web projects run independently from 2005 to 2011, and the first sites delivered to clients.
Personal project launched in September 2026: a showcase site backed by a professional career database, published in French, English and Spanish, and generated entirely ahead of time. The same database feeds a second project that generates CVs and cover letters.
Role held: AI Product Manager. Technology stack selection — static generation, a managed PostgreSQL database, an MCP server as the only write path, a private server and Caddy —, decisions on the data model and privacy rules, and setup of the AI-assisted development environment: project rules, skills, review agents specialised in eco-design and accessibility, and hooks that enforce guardrails rather than merely recommending them. The code is written with Claude Code; delivery is steered through a Linear backlog of 155 tickets in five weeks.
The site is meant as evidence: no figure is published without a measurement. Weight, EcoIndex and Lighthouse scores are measured in production, and an accessibility and eco-design audit follows every deployment, automatically opening a ticket at the slightest deviation. Two checks block publication: a secret scan and a verification that client names stay anonymised.
Achievements
- ~700 pages generated in three languages
Fully static, trilingual site
Every page is produced ahead of time and served as static HTML, in French, English and Spanish. A page without a translation is not generated in that language: no fallback to French.
- 0 JavaScript file
Site served without any JavaScript
No JavaScript file is produced or delivered: reading, navigation and language switching rely on HTML and CSS alone.
- 8 kilobytes at most for the heaviest page
Weight budget held on every page
Bytes actually transferred, compression on, sub-resources included, measured in production by a script in the repository and never typed in by hand. The heaviest page is the skills catalogue.
- 9 pages graded A out of the 9 in the statement
EcoIndex measurement of every page in the statement
Scores from 81 to 93 out of 100, measured with lighthouse-ecoindex and checked against the published figures after every deployment. The skills catalogue, first graded B, moved to A by listing only skills tied to at least two assignments. Lighthouse performance and accessibility scores at 100.
- 100 % of applicable RGAA criteria met, 41 out of 41
RGAA 4.1.2 self-assessment across 13 pages
Four non-compliances found in the first pass — table headers, unflagged English terms, information conveyed by formatting alone, hidden focus — fixed, then checked in a second pass. Accessibility statement published in three languages.
Automated audit after every deployment
Every publication triggers an audit: contrast in both themes, visible focus on every keyboard stop, display at 320 px and enlarged text, HTML validity, EcoIndex and published figures. The slightest deviation opens a ticket; every blocking check was first seen to fail.
- 4 skills, 1 of them built for the project
Setup of the agentic development environment
Project rules, two specialised review agents — eco-design and accessibility — and two hooks that enforce guardrails: an impact reminder on the database shared with another project, and read-only Bash for the reviewers.
Skills applied
- 150 web eco-design best practices
- Agentic development environment setup
- AI Product Management
- Building skills for AI agents
- Claude Code
- Context engineering
- Continuous integration and deployment (CI/CD)
- Corrective action planning after audit
- Data migration
- Designing MCP tools
- Designing specialised AI agents
- Digital accessibility
- Eco-design assessment
- Eco-design lead for a digital service
- Eco-design measurement tooling
- EcoIndex
- Functional acceptance testing and qualification
- GR491
- Guardrails through AI agent hooks
- Internationalisation (i18n)
- Lighthouse
- Linear
- Product ownership
- Product roadmap
- Project management
- Relational data modelling
- RGAA
- RGESN
- Security vulnerability remediation
- SEO
- Supabase
- Technology stack selection