Full-stack JavaScript engineer. Node/Bun back ends, Nuxt front ends, and the Docker Swarm and k3s clusters they actually run on. Seven years of JavaScript, working solo — which means I own the whole chain, from the query to the 3am incident.

My public work lives in two repositories: resilient-ws — a dependency-free WebSocket client that detects the socket which stays open and silently stops delivering, extracted from my own ingestion stack. agent-skills — production knowledge distilled from real postmortems, packaged for AI coding agents. Every rule in it cost me hours or days to learn.

WebSocket ingestion that survives production. I ingest live trades from about ten crypto exchanges. Connecting was never the hard part — the hard part is the socket that stays open while silently delivering nothing. Application-level ping/pong on top of the transport, zombie detection, clean reconnect with no orphaned timers. I extracted that invariant and published it: resilient-ws — including the case where the rule inverts, and why CCXT Pro does not catch this failure.

Data at scale, and the ways it breaks. MongoDB in depth: index-first with .explain(), grouped writes, the TTL that silently no-ops because the field isn't a Date, the engine cache that fights the container's memory cap. I've paid for each of those in production.

Infrastructure I run myself. Docker Swarm, then a single-node k3s cluster in real production — ingress reachable only through a WireGuard tunnel, CronJob backups, a dead-man's switch, reboot-durability tested live. I host what I ship and keep it alive.

Vue and Nuxt, plus Three.js when the interface needs 3D — including the Blender side: glTF import, cleanup and re-export, Draco compression, and why your scene is slow. I shipped a browser game for a paying client.

AWS, GCP, Firebase, Supabase. Managed cloud isn't where I work, and I'd rather tell you that now than learn it on your budget.

• A MongoDB TTL index on Date.now() does nothing. No error, no log. 127 GB and two months of work, gone. Not my first scar — but a good one.
• kubectl top under-reports by a factor of three. I went looking for headroom that was never there.
• 137 restarts in 11 hours. Memory was fine. It wasn't an OOM — my own liveness probe was killing the service because it couldn't answer /health under batch load. Another week gone for my delight.

All of it, and the rest, is in agent-skills and resilient-ws.

Based in France. I work in English and French — I write English a good deal better than I speak it, so written and async collaboration suits me best.

Available for freelance work, remote. Back ends and ingestion pipelines, MongoDB, self-hosted infrastructure. The write-up behind resilient-ws is here.

Reach me at carlbouvetdev@gmail.com or on LinkedIn.