Notes
Short essays, cross-posts, and technical notes.
-
Gmeow: your mailbox is agent memory, keep it local
The recurring problem with useful personal agents is that the context they need already exists, but not in the tidy place a demo would like it to be. It is in mail. gmeow makes that mailbox substrate locally useful through loopback REST, MCP, PostgreSQL, semantic search, attachment sidecars, object storage, and a local knowledge graph.
-
Agentic coding needs human sign-off tied to physical reality
As AI coding agents transition from passive autocomplete tools to autonomous contributors executing entire feature branches, we are racing toward a massive security blind spot: how do we prove a real human actually reviewed and verified agent-generated code before it hits production?
-
Engineering principles should be runnable policy, not slide-deck nostalgia
If your team's standards live in a slide deck, your AI agents will violate them. coding-ethos compiles one YAML file into linter configs, git hooks, agent prompts, and an MCP server so the rules cannot drift between human and machine readers.
-
Standard Graph Neural Networks need curved semantic manifolds
Standard GNNs are structurally constrained when mapping complex text attribution; linear aggregation in flat Euclidean space inevitably forces semantic drift. Looking to connect with researchers in TDA, geometric deep learning, and spectral graph theory.