ARCHITECTURE SPECIFICATION: Expedify Agentic Content Engine

1. Project Overview & Business Context

Project: Autonomous SEO Content Engine for Expedify ([https://expedify.com/en/](https://expedify.com/en/))

Business Model: Expedify is a specialized processor for international travel documents (Visas, International Driving Documents, Passports, and Apostilles/Civil Records). They handle the bureaucratic friction of government paperwork, passing official government fees through at cost.

Domain Classification: Strict YMYL (Your Money or Your Life).

The Core Problem: Why Standard AI SEO Fails Here

A generic AI content engine will fail catastrophically in this domain. If an LLM relies on its internal weights to write about visa requirements, it will hallucinate outdated government fees, confuse the 1949 Geneva and 1968 Vienna driving conventions, or invent processing times. In this vertical, a factual error doesn't just result in a high bounce rate—it results in a customer being denied boarding at an airport.

Therefore, this architecture must strictly decouple hard data from narrative generation, run comprehensive legacy audits to prevent cannibalization, and enforce rigorous guardrails before publishing.

2. Orchestration Framework: Hermes Agent

This system will be orchestrated using Hermes Agent (by Nous Research) rather than manual workflow builders like n8n. The builder should leverage Hermes' native capabilities:

Bot Mode / Subagents: Deploy specialized, isolated subagents (Bots) for specific pipelines (e.g., a dedicated YMYL_Compliance_Bot, a Semantic_Mapper_Bot).

Python RPC Scripts: Use Hermes' programmatic execute_code and RPC scripts to collapse multi-step API pipelines (like schema injection) into zero-context-cost operations.

MCP Integration: Connect Hermes to your CMS and structured databases via MCP (Model Context Protocol) servers to read legacy URLs and write finished JSON-LD.

Scheduled Automations: Utilize Hermes' built-in cron scheduler for the Content Decay and Visibility pipelines.

3. The 10-Pipeline System Architecture

Phase 1: The Knowledge & Strategy Layer

Ensures the engine integrates with existing content, establishes ground truth, and prevents hallucinations.

Pipeline 1: The Legacy Audit & Entity Reconciliation (Run First)

Function: An exploratory Hermes subagent crawls the current Expedify sitemap, vectorizes existing guides (e.g., "Driving in Japan"), and extracts their core entities and conventions.

Requirement: It builds a master URL registry mapping existing pages to their semantic intent, tagging them to be kept, updated, or redirected. This prevents the system from generating duplicate hubs that cannibalize current rankings.

Pipeline 2: The Semantic Graph & Topical Hub Orchestration

Function: Using Hermes' Firecrawl integration, the agent scrapes top SERP results to map the semantic relationships (triples) competitors use to satisfy search intent. It structures the hub-and-spoke hierarchy before any content is drafted.

Requirement: The builder must strictly adhere to the entity structures and semantic rules defined at [https://www.bigtechies.com/semantics/](https://www.bigtechies.com/semantics/) to ensure the ontology remains clean.

Pipeline 3: Programmatic SEO (pSEO) Data-Sync (Critical)

Function: Connects to official government portals to ingest hard data (embassy fees, processing times, required treaties) into a structured database via MCP.

Requirement: Content generation bots must never invent facts; they are only permitted to pull and inject variables from this synchronized database into the semantic templates.

Phase 2: The Generation & Quality Layer

Drafts the localized content and enforces strict YMYL compliance.

Pipeline 4: RAG-Powered QA & Information Gain

Function: Evaluates the draft against the live SERP to guarantee Information Gain (unique perspectives, proprietary Expedify workflow steps). Drafts are checked against a chunked/vectorized internal knowledge base of Expedify's actual SOPs.

Pipeline 5: YMYL / E-E-A-T Validation (Hard Gate)

Function: A rigid compliance Bot scans the final draft.

Requirement: If a claim regarding a legal requirement lacks a direct citation/link to an authoritative government source (.gov, official consulate), the Bot must fail the build and kick it back. It also automatically injects the appropriate expert "Person" entity attribution.

Pipeline 6: Multi-Regional Semantic Localization

Function: Translates the core semantic triples into local search intent for non-English markets, ensuring cultural and terminological accuracy (e.g., Japanese specific intent for driving permits) rather than doing literal 1:1 translations. It also maps cross-linking hreflang architecture.

Phase 3: The Technical & Structural Layer

Wires the machine-readable architecture for search engines.

Pipeline 7: Dynamic Internal Link & Hub Orchestration

Function: Governs internal linking to prevent intent cannibalization, utilizing the registry built in Pipeline 1.

Requirement: Rigidly silos informational destination guides (e.g., "Driving in France") from transactional application hubs, ensuring PageRank flows correctly to converting pages.

Pipeline 8: Technical SEO & Dynamic Schema

Function: A technical subagent dynamically constructs the exact JSON-LD structured data needed.

Requirement: Uses Hermes RPC scripts to push the schema directly to the CMS, tying Article schema to verified Person (the expert reviewer) and Organization (Expedify) entities.

Phase 4: The Defense & Lifecycle Layer

Actively defends rankings and adapts to search volatility post-publication.

Pipeline 9: Content Decay & Lifecycle Automation

Function: Using Hermes' built-in cron scheduling, an agent monitors the Google Search Console API.

Requirement: If an established page drops out of the top 3, or if Pipeline 3 detects a government fee change, the agent autonomously analyzes the shift and outputs a highly specific refresh brief.

Pipeline 10: AI Search (GEO) Visibility Tracking

Function: Periodically queries foundational models (Perplexity, AI Overviews) via the Tool Gateway to monitor brand visibility for key queries (e.g., "Do I need a visa for Japan?").

Requirement: Generates mention gap reports to inform off-page strategy and entity association in LLM training sets.

4. Hermes Implementation Notes for the Builder

Memory System: Leverage Hermes' persistent memory across sessions to ensure the agent remembers the Expedify brand voice and previously mapped legacy URLs without needing to re-ingest them every run.

Zero-Shot Prohibition: Apply custom system prompts to the generation subagents explicitly forbidding them from relying on their pre-trained weights for fees, dates, or legal requirements.

Human-in-the-Loop (HITL): For Pipeline 5 (YMYL Validation), use Hermes' messaging gateway (e.g., Telegram or Discord integration) to ping a human editor with the final draft and YMYL compliance score. The agent must pause execution and wait for human /approve before executing the CMS publish script.

This document is the original requested architecture specification. The canonical build plan and salvage audit may refine its implementation order and governance while preserving its core requirements.