I'm a software engineer. I work on multi-agent systems and full-stack apps, and I write here about what I learn while building them.
Efficient RAG is not a better embedding model. It is chunking that retrieves one idea at a time, plus an incremental ingest pipeline that upserts changed chunks by deterministic IDs and deletes the rest.
Standard RAG finds similar text. Graph engineering walks the chain of causes. Here is the eight-layer architecture I would actually build, and why long context is the right engine for it.
Plugins are how you ship rules, skills, agents, MCP, and hooks as one bundle. Here's the structure, local workflow, and practices that keep agent context lean.
I'll email you when I publish something new. Mostly AI agents, frontend, and things I'm learning on the job. Unsubscribe any time.