Technical deep dives on AI systems, architecture patterns, and tools I'm building. Written to be read, shared, and referenced.
Trust-scored long-term memory, self-correction rules, Obsidian vault as knowledge store, multi-agent shared memory, and overnight research sprints.
How to build a structured personal knowledge system in Obsidian with freshness tracking, canonical authority, cross-cutting tags, and AI-assisted maintenance.
How to organize a personal knowledge system across Google Drive, Obsidian, and your local filesystem. What goes where, why they mirror each other, and how AI agents read all three.
Generic embeddings can't tell Spanish tax consultations apart. Three approaches to fixing retrieval on 17,736 legal documents, and how autonomous overnight fine-tuning on Apple Silicon is changing things.
How OpenClaw routes messages through iMessage via the Beeper bridge. Sequence diagrams, relay architecture, TTS pipeline, and the full message lifecycle.
Council's architecture: 7 calculation engines, knowledge packs with 2,100+ verified sources, deterministic math, and the trust chain from government publications to client reports.
A comparative case study on domain-specific governance emergence in AI-assisted decision systems.