Senior software engineer, 20+ years. Available immediately; US Pacific hours. Looking for a full-time remote role designing backend or platform systems.
I write programs that write programs. Data UI is a compiler I'm building in Common Lisp. It already compiles a 57-line model into PostgreSQL, REST, a minimal React UI, row-level RBAC, and Kubernetes+TLS, and deploys it. The same application is roughly 20,000 lines written by hand. In my tests an agent can generate large parts of it, but not the application. It needs a senior directing it the whole way. The 57-line model is what gets reviewed instead. Live demos: <a href="https://data-ui.com" rel="nofollow">https://data-ui.com
Recently, Azure Government (FedRAMP IL5): commercial Elasticsearch was not allowed, so I replaced it with self-hosted Elasticsearch inside the VNet. Subnet and AKS in Terraform; Elasticsearch from Terraform-managed Helm charts.
I run several models from different providers, including local ones, in one Emacs instance. Any of them can enlist another against the same live state. One shape: a model drafts a plan, enlists a second to draft its own, then reconciles the two. They implement against the live Lisp image, and they can patch the Emacs instance they are running in, so a fix to the harness takes effect on the next prompt. In the usual setup the model can't reach the program running it. Only I commit.
macnod · · focus · HN ↗
Remote: Yes (required)
Willing to relocate: No
Technologies: Python, Java, SQL, AWS, Azure, Kubernetes, Terraform, PostgreSQL, GitHub Actions, Common Lisp
Résumé/CV: <a href="https://sinistercode.com/public/donnie/resume" rel="nofollow">https://sinistercode.com/public/donnie/resume
Email: macnod [at] gmail [dot] com
Senior software engineer, 20+ years. Available immediately; US Pacific hours. Looking for a full-time remote role designing backend or platform systems.
I write programs that write programs. Data UI is a compiler I'm building in Common Lisp. It already compiles a 57-line model into PostgreSQL, REST, a minimal React UI, row-level RBAC, and Kubernetes+TLS, and deploys it. The same application is roughly 20,000 lines written by hand. In my tests an agent can generate large parts of it, but not the application. It needs a senior directing it the whole way. The 57-line model is what gets reviewed instead. Live demos: <a href="https://data-ui.com" rel="nofollow">https://data-ui.com
Recently, Azure Government (FedRAMP IL5): commercial Elasticsearch was not allowed, so I replaced it with self-hosted Elasticsearch inside the VNet. Subnet and AKS in Terraform; Elasticsearch from Terraform-managed Helm charts.
I run several models from different providers, including local ones, in one Emacs instance. Any of them can enlist another against the same live state. One shape: a model drafts a plan, enlists a second to draft its own, then reconciles the two. They implement against the live Lisp image, and they can patch the Emacs instance they are running in, so a fix to the harness takes effect on the next prompt. In the usual setup the model can't reach the program running it. Only I commit.
Portfolio: <a href="https://sinistercode.com/public/donnie" rel="nofollow">https://sinistercode.com/public/donnie
GitHub: <a href="https://github.com/macnod" rel="nofollow">https://github.com/macnod