Field CTO at Siel AI and a cloud and AI architect. On GitHub I build open-source tools for the work around enterprise AI: FDEOps for AI coding agents, Awesome Enterprise AI for business workflows and CodeQuorum for pull request review.
I like figuring out what is worth building in the first place. Design thinking keeps me close to the people facing a problem, systems thinking shows how a fix fits the bigger picture, and first principles help when the usual approach doesn't make sense. FDEOps is that method written down as skills. I look for a small, useful place to start on big problems. In my own time, I use GitHub to build things developers and enterprises can try, question and improve.
947 stars · 106 forks · 6,500 npm downloads in September · CI · Evals · 51 releases
Skills for AI coding agents, from scoping a problem through verification and handover. FDEOps keeps a local record of decisions so the next session can pick up the work. It installs into Claude Code, Cursor, Codex and Copilot.
The fieldbook, a read-only view of local customer records. Run npx fdeops demo to generate it from sample notes.
Python · 153 tests, a 58-case live evaluation and 20/20 on independently written cases · Try the workspace · MIT
Small workflows for the work around enterprise AI: investigating incidents, checking claim evidence, answering business questions and coordinating operations. Each includes examples, tests and a governance file that names the business owner, data risks and the thresholds to agree before adoption. Nine are reviewed references; the proposal-evidence project remains experimental. RAG Scope Check, a local CI gate for permission-scoped retrieval, is also available on its own; it started from a practitioner's report that access filtering was quietly starving retrieval.
Python · FastAPI · Docker · CI on every push · Scaling limits · MIT
A reference architecture for generating AI video in batches. The interesting part for me is the workflow around generation: routing jobs between models, retrying failures and tracking cost. The README documents the operating limits and what would need to change for larger runs.
Python · LangGraph · GitHub Action · Tests · MIT
Three agents review a pull request with different priorities. Findings backed by at least two reviewers are marked "fix it"; a human can inspect the reasoning in the review comment. Agreement is a signal to investigate, not a guarantee that the finding is correct.
Merged fixes in OpenClaw: Slack reconnects, Telegram approvals, Ollama model IDs, OpenAI reasoning blocks and hook context.
- Multi-Cloud Handbook for Developers (Packt), a book on designing and running cloud-native applications across AWS, Azure and GCP, co-written with Jeveen Jacob.
- Agentic AI Architecture Framework for Enterprises on InfoQ, co-written with Ahilan Ponnusamy.
- Articles on The New Stack about multicloud strategy and FinOps.
- FDE field notes, a newsletter on forward deployed engineering.
If you try one of these tools, I'd like to hear where it helps and where it falls short. Reproducible problems and small contributions are welcome.