Cutting Edge Tech

What I build so coding agents stay oriented, honest, and on task.

Featured2026

Enforcement Hooks

Agents that grade their own work ship drift. Two hooks (line-level on every write, intent-level at every Stop) keep the agent in lock step with the project's standards.

Read the case study

Sylphie2026

Cognitive architecture

A non-LLM-in-the-runtime-path cognitive architecture. Dual-process reasoning, drive-driven attention, language as voice box rather than thinker.

Read the case study

memory-pkgnpm · memory-pkg2026

Agent long-term memory

Agents forget the work between sessions. memory-pkg indexes every transcript into TimescaleDB and trigram-searches the right history into the next prompt. Pure SQL, no LLM in the hot path.

Read the case study

sylphie-pkgnpm · codebase-pkg2026

Codebase knowledge graph

Coding agents waste context re-reading files. sylphie-pkg compiles the source tree into a Neo4j graph and serves it over MCP: a map for the agent, not a flashlight.

Read the case study

The Missing Layer: Why Procedural Knowledge Graphs Will Reshape AI Coding2026

Structure for agent code decisions

Procedural knowledge graphs separate what an AI model should decide (what to build, how to decompose it) from what it shouldn't (the actual code shape). By encoding how your team writes code, a PKG lets the model focus on interpretation and planning while a deterministic graph handles code generation.

Read the case study