I spend most of my time making systems agree on data they were never meant to share. EDI, ETL, APIs and workflow orchestration with n8n.
On the side I build ControlDeck, a learning project where a few AI agents work together on a real codebase. Rule of thumb: deterministic pipelines first, LLMs only where they earn their place.
- Pipelines that fail loudly and retry safely: idempotency, backoff, clear error paths
- Data formats that actually get used: EDIFACT, XML, JSON, CSV
- Agentic coding with real guardrails, reviews and tests
Python SQL TypeScript n8n REST Git Linux
More thoughts on X: @flkcgn


