HUGO.CHARMIES(1)
PRODUCTION
$ hugo --whoami
Full stack developer in Belgium. I build internal tools that replace manual and paper-based work, most recently at Vanwezel Autoparts.
$ hugo --status
status : ● Open to full stack roles location : Limburg · Belgium daily_driver : Claude Code favourite_stack: Next.js · Postgres
$
$ hugo --now
$ hugo --featured








$ hugo --stack
AI
────────────
FULL STACK
────────────
INFRA & TOOLS
────────────
$ hugo --origin
I studied Applied Computer Science in Leuven. For a school project, a client needed indicators scraped from the web, which was all manual work. I automated it with an AI model and saw how much faster you can build now.
Since then I build with AI coding tools like Claude Code, and I test and review what they produce. What I like most is finding a slow process and replacing it with a small app.
EDUCATION
UC Leuven Limburg
BSc Applied Computer Science · 2022–2025
FIRST AI PROJECT
Ecofoodmap · n8n → code-native
WAY OF WORKING
Claude Code, then test and review
$ hugo --archive
$ hugo --contact
$ open ~/projects/charmies
HUGO.CHARMIES(1)
STACK
Next.js · Supabase · Vercel AI SDK · tool calls · RAG
01 / 04
The brief. Forty years of staffing run on Excel sheets, Facebook groups, and Word contracts over WhatsApp. Collapse it into one platform, and make the next layer AI-native, not a chatbot bolt-on.
02 / 04
What's actually live. A production agent that replaces manual creation flows: "schedule three charmies for tomorrow's brand activation" triggers the agent, the agent calls the tools, a confirmation card lands. Built on the Vercel AI SDK with structured tool calls + a RAG layer over the agency's own docs. About 200 ambassadors and the 5-person ops team use it daily. Contract issuance, the agency's slowest workflow, went from 10 to 15 minutes per contract to under one, end to end.
03 / 04
Trust, but verify the agent. A real B2B app is where AI-generated code fails quietly: IDOR, RLS, multi-tenancy bleed-through, GDPR edge cases. Most AI-built portfolios skip this layer because the model doesn't surface it. I wrote the e2e suite explicitly around those failure modes: server-side JWT role guards, IDOR coverage across 4 role surfaces, race conditions, GDPR paths, verified before each merge. This is the discipline AI codegen actually needs.
04 / 04
The arc that matters. I started this dashboard when my agentic discipline was weaker. Early shipped code was messy, fixes broke other things. The last six months were a stabilisation pass: surgical fixes on a large existing AI-built codebase, no rewrite, while the agency kept using it. Managing the downside of AI codegen is the 2026 skill. Most engineers can ship with Claude. Fewer can clean up after themselves.
SCREENSHOTS
$ Sole engineer end to end, architecture, deployment, the agent, the security envelope. Next: expanding the agentic surface to the client-facing booking flow.