AI-augmented developer in Paris. Python automation, data extraction, full-stack MVPs — delivered asynchronously, with the production infrastructure to back it up.
I build the machinery that moves data — and the guardrails that keep it honest when nobody is watching.
Background: data science and full-stack development, now working as a solo AI-augmented builder. Day to day that means Python automation, web scraping (including sources that fight back), structured extraction from PDFs and reports, data pipelines into Sheets/Postgres, API integrations, and full-stack MVPs in Next.js / React / FastAPI when the data needs an interface. LLM integration — Claude, OpenAI, RAG — runs through most of it.
How I work is the part worth knowing. I ship with AI coding agents in the loop, and I have open-sourced the infrastructure that makes that survivable in production:
- claude-code-cron — unattended scheduled runs, git-push distributed lock across machines, failure alerting
- claude-code-safety-hooks — guardrail hooks that gate dangerous shell commands and block hardcoded secrets before they reach a commit
- claude-code-todo-digest — nightly cross-repo digest via GitHub Actions, zero LLM calls, pure regex
github.com/yvoolab — all four repositories are mine, none are forks.
Why that matters to you: most automation fails quietly. A scraper that returns zero rows looks exactly like a day with no data; a scheduled job that dies at 3am looks exactly like a job with nothing to do. A large share of what I build exists purely to tell those two apart — because the failure modes, not the happy path, are what decide whether an automation is worth having.
Based in Paris (CET). English, French, Chinese. Registered French micro-entreprise, invoicing compliant.
Work Terms
Communication — Asynchronous by default. I work from written briefs and I answer within one business day. I do not need standing meetings to make progress, and I would rather send you a working prototype than a status update. If a call is genuinely faster, one short call is fine; a recurring meeting series is not how I work.
Hours — Paris time (CET/CEST). I do not commit to being online during any specific hours, including US working hours. Work is delivered on schedule, not supervised in real time.
Pricing — Fixed price per deliverable is my strong preference: you know the cost up front, and my speed works for you rather than against you. Hourly is available where a project genuinely calls for it. Scope changes are re-quoted before work continues, never billed as a surprise.
Payment — Through Guru SafePay. Milestones for larger projects: an agreed portion released at each defined deliverable.
What I need from you — A clear description of the outcome you want, and the data or system details the work depends on. I will tell you plainly if something is outside what I can deliver well, before you pay for it rather than after.
What I will not do — Anything requiring me to appear on camera or be present in person, content designed to mislead, or work where the deliverable is my presence rather than my output.