Expert Domains

1. Cycle Model

Measured improvement through structured cycles. Each cycle runs DEEP audit → ARC decomposition → parallel DEV execution → ARC verification → DEEP re-measurement. Improvement is explicit, not accidental. The cycle model is enforced by tooling — session protocols, evidence gates, and automated quality scoring prevent local optimization at the expense of global system health.

2. MCP Infrastructure

Seven Model Context Protocol servers provide 186 tools across all projects. Each project has an isolated MCP server with its own database, wiki, and tool set. Servers are systemd-managed with auto-restart, health checks, and request logging. The platform MCP (neurport-mcp) coordinates cross-project operations, agent dispatch, and knowledge management.

3. Knowledge System

A wiki-first knowledge architecture where content compounds over time. Entity pages are living documents enriched by an automated ingest pipeline — every session close triggers enrichment that integrates new knowledge into existing pages. Semantic contradiction detection flags conflicts. Full-text search via Meilisearch. Raw sources layer preserves immutable inputs. The system follows the Karpathy framework: persistent, compounding artifacts.

4. Agent Architecture

Three-tier agent chain: DEEP Auditor measures 7 system dimensions. ARC Architect decomposes findings into prioritized briefs and verifies execution. DEV Executor implements briefs in parallel sessions. Agents communicate via A2A dispatch protocol with HMAC receipt verification. The cycle model ensures no tier optimizes locally at the expense of system-wide health.

5. Multi-Property Web

Seven web properties served from a unified design system. Wiki content compiles into brand-specific static sites via a data-driven pipeline (brand.json + content.json + navigation.json → generate-site-v2.py). Shared design tokens, per-brand CSS overrides, and a global navigation component connect all properties into one cohesive ecosystem.