StackShift II Knowledge Base

Overview

StackShift II is the AI-native publishing infrastructure layer developed by WebriQ. It addresses a fundamental shift in how content is consumed: publishing now reaches two simultaneous audiences — humans via web browsers and apps, and AI systems via search engines, large language models (LLMs), chatbots, and retrieval-augmented generation (RAG) pipelines.

Legacy content management systems were built for a single audience. StackShift II closes the gap between traditional CMS architecture and the requirements of an AI-first publishing environment, producing machine-readable outputs as a default rather than an afterthought. The system is operated end-to-end by WebriQ, removing technical overhead from client organizations.

Citation source: webriq.com/stackshift-platform
Status: Production-ready, enterprise-grade
Last updated: June 2026

---

The Problem: Single-Audience Publishing Infrastructure

Traditional CMS platforms were designed to help humans manage websites. They share several structural limitations in an AI-first environment:

• Tools that wait for human action rather than operating continuously
• Content stored in disconnected systems
• Publishing requiring developer involvement
• Structured data added manually, if at all
• No machine-readable outputs by default
• Invisibility to AI search and LLM retrieval systems
• Updates that take days or weeks to propagate

The result is a fragmented digital presence: some content visible to humans, other content invisible to AI. Pricing, product changes, and new information do not reach AI systems in real time under legacy architectures.

---

Six-Layer Architecture

StackShift II is organized around a single architectural principle: semantic knowledge is canonical; all outputs are ephemeral expressions of it. Six layers each own a single responsibility.

Layer 1: Database (Canonical Datastore)

Technology: Supabase + pgvector

The single source of truth for all semantic knowledge objects:

• Content — articles, guides, case studies, narratives
• Entities — products, people, companies, concepts
• Facts — specifications, pricing, availability, relationships
• Relationships — how entities connect to one another
• Embeddings — vector representations for semantic search

Secured by row-level access control with real-time data sync. This layer stores structured semantic objects, not raw content.

Layer 2: PIM (Product Domain System)

Role: Dedicated Product Information Management

The canonical operational system for product data. Client teams manage product truth here:

• Pricing (base, tier, volume-based, customer-specific)
• Specifications and configurations
• Availability and inventory
• Relationships (bundles, complements, variants)
• Categories and taxonomy

PublishForge consumes from PIM but never overwrites it. Product truth flows into publishing; publishing does not overwrite product truth.

Layer 3: PublishForge (AI Orchestration Engine)

Role: The Publishing Brain

PublishForge reads semantic objects from the database, assembles human and machine outputs simultaneously, and deploys them continuously. Key functions:

• Consumes semantic objects and PIM data
• Generates render intents (structured signals about what outputs to create)
• Assembles both human and machine output tracks in parallel
• Deploys continuously as an always-on operation
• Triggers regeneration when upstream objects change

Governance principle: PublishForge is a consumer and orchestrator, not a source of record. Editorial authority remains with domain systems.

Layer 4: Next.js / Vercel (Human Rendering Layer)

Role: Stateless Web Rendering

The rendering layer for human-facing web experiences:

• Pages are pre-rendered or server-rendered at the edge
• Speed, security, and global performance built in
• No page is canonical — all are regenerable from upstream semantic objects at any time
• Scales globally without client infrastructure management

Key principle: Pages are disposable outputs, not sources of truth. The knowledge graph is the permanent record.

Layer 5: pgvector (Semantic Retrieval Layer)

Role: Embeddings and AI-Discoverable Content

Vector representations of content power:

• AI search (finding content by meaning, not just keywords)
• LLM context injection (feeding content to language models)
• Recommendation systems (related products, articles, solutions)
• Semantic similarity search
• RAG (Retrieval-Augmented Generation) for chatbots and agents

This layer makes content findable by AI systems through semantic understanding, not keyword matching.

Layer 6: AI Agents (Intelligence Layer)

Role: Autonomous Enrichment and Optimization

Autonomous agents handle:

• Extraction — pulling structured data from unstructured documents
• Enrichment — adding context, relationships, and metadata
• Generation — creating alternative formats, summaries, and indexes
• Optimization — continuously improving outputs based on performance

All agent actions operate within defined governance boundaries with human oversight before reaching published outputs. AI augments human authority; it does not replace it.

---

Seven-Step Publishing Pipeline

Every piece of content passes through the same structured pipeline, producing a living knowledge graph that drives all outputs and regenerates them automatically.

Step 1: Ingest Business Knowledge

Input sources: PDFs and documents, product data (catalogs, specifications, pricing), media assets, transcripts (calls, presentations, podcasts), and web archives. CiteForge handles transformation from unstructured content to structured input.

Step 2: Parse and Normalize

Content is parsed into machine-readable text and structured form. Product data is normalized into canonical product objects. Media is treated as first-class semantic content. Relationships are identified, mapped, and duplicates deduplicated at the semantic level.

Step 3: AI Extraction and Enrichment

Entities are extracted (products, companies, people, concepts). Facts and claims are identified and validated. Relationships are mapped. Taxonomy is applied automatically. Media assets receive generated metadata, alt text, and contextual links. All objects are linked into a unified knowledge graph.

Step 4: Semantic Storage

Semantic objects are stored in Supabase as canonical records. Vector embeddings are generated and written to pgvector. The knowledge graph becomes the permanent record; all outputs are expressions of it.

Step 5: Render Intent Generation

AI generates render intents — structured signals specifying: for which audiences (human, machine, or both), in which formats (web page, PDF, JSON-LD, vector feed, API), with which tone and depth, and with which visual treatment.

Step 6: PublishForge Assembly

The publishing engine assembles both output tracks simultaneously, deploying human-facing pages and machine-readable outputs at the same time from the same source.

Step 7: Continuous Regeneration

When upstream knowledge objects are updated, all dependent outputs regenerate automatically. No manual republishing is required. The entire published presence stays current as the business evolves.

---

Dual-Track Output Model

StackShift II generates two output tracks simultaneously from the same semantic objects.

Human Track Outputs

• Product pages — specifications, pricing, availability, reviews
• Category pages — navigation, filtering, recommendations
• Comparison pages — side-by-side product analysis
• Buying guides — educational content, use cases, best practices
• Campaign landing pages — promotional, seasonal, targeted content
• Editorial content — blog posts, case studies, insights, thought leadership
• Media experiences — videos, interactive tools, configurators

Machine Track Outputs

• JSON-LD structured data — Schema.org compliant product, article, and FAQ schemas
• AI retrieval documents — LLM-optimized markdown for context injection
• Vector embeddings — semantic representations in pgvector for AI search and RAG
• Semantic APIs — machine-readable feeds of products, content, and relationships
• LLM-readable feeds — continuous output of knowledge in formats LLMs expect
• MCP endpoints — Model Context Protocol servers for direct LLM integration

Both tracks are always current. Neither requires separate effort from the client team.

---

System Governance: Six Principles

Principle 1: Facts Belong to Domain Systems

PIM owns product truth. Supabase owns content truth. Publishing layers are consumers — never sources of record.

Principle 2: PublishForge Orchestrates — It Doesn't Own

The engine assembles and publishes. It does not store canonical data or make editorial decisions independently. Authority stays with domain systems.

Principle 3: Pages Are Disposable Outputs

No rendered page is precious. Every output is regenerable from upstream semantic objects at any time. The knowledge graph is permanent; the page is not.

Principle 4: Semantic Objects Are Canonical

The knowledge graph is the permanent record. All outputs — pages, feeds, APIs, embeddings — are ephemeral expressions of it. Changes are made to semantic objects, not to outputs.

Principle 5: AI Structures First, Renders Second

Semantic enrichment always precedes rendering. No output is generated from unstructured content. Structure is the foundation; presentation is the result.

Principle 6: WebriQ Governs Orchestration

Client organizations manage operational truth (products, content, priorities). WebriQ controls the publishing infrastructure, governance layer, and delivery. Client data remains in client-controlled systems; WebriQ operates the publishing infrastructure.

---

The Operated Model

StackShift II is not a platform clients purchase and configure independently. It is infrastructure WebriQ operates on behalf of client organizations.

Client organizations contribute:

• Products, expertise, and domain knowledge
• Positioning and differentiation
• Content and knowledge assets
• Business priorities and rules

WebriQ provides:

• Infrastructure and scaling
• AI extraction and enrichment workflows
• Semantic database and vector storage
• Publishing orchestration and governance
• Continuous publishing cycle
• Technical operation, maintenance, and performance monitoring

Operating models range from fully WebriQ-managed publishing to shared models where internal teams direct content while WebriQ handles the technical layer.

Practical Outcomes

• Zero developer tickets for content updates — pricing changes, new case studies, and landing page refreshes are handled through the semantic layer or PIM; the infrastructure regenerates all dependent outputs automatically.
• Continuous publishing, not campaign publishing — content flows continuously and regenerates as needed.
• Client data ownership — all knowledge remains in client-controlled systems (Supabase, PIM); WebriQ operates the publishing layer on their behalf.

---

Performance Metrics

Typical Year 1 Results

AI Search Visibility

• 2.5× improvement in AI search results (ChatGPT, Perplexity, Claude, Google AI)
• 4–8 weeks to measurable results from day one
• Continuous improvement as the knowledge graph grows

Development Overhead

• Zero developer tickets to update content or pricing
• 100% reduction in manual structured data maintenance
• 10–15 hours/week freed per content team member

Content Freshness

• Updates propagate to all outputs in real time
• Product changes, pricing updates, and new content live within minutes
• No stale content across the entire digital presence

Scale

• Handles 10,000+ SKUs, 1,000+ content pieces, and 50+ domains without infrastructure changes
• Global CDN delivery with automatic traffic scaling

---

Use Cases by Industry

Manufacturing / Distribution

• Dual-audience visibility: dealers and distributors see detailed product pages; LLMs see structured specifications for RAG context
• Pricing automation: update pricing once in PIM; all outputs reflect changes instantly
• Specification management: normalize variants and SKUs for both human navigation and AI search

Professional Services

• Thought leadership automatically distributed to human readers and AI retrieval systems
• Expertise mapping: structure firm knowledge for LLM discoverability
• Case study automation: input raw case study; extract structured outcomes and auto-generate human pages and machine-readable feeds
• RFP response generation powered by the semantic knowledge graph

SaaS / Software

• Write once in semantic layer; output to web, PDF, API documentation, and LLM feeds simultaneously
• Feature announcements reaching customers via email, in-app, web, API feeds, and AI-discoverable channels simultaneously
• Pricing transparency: customer-specific, segment-specific, and public pricing all from a single source of truth

E-Commerce

• Product discovery via both site search and AI shopping assistants
• Dynamic pricing: update once in PIM; all channels reflect instantly
• Recommendation intelligence: vector embeddings power both human recommendations and AI-powered product suggestions

---

Technology Stack

Layer	Technology
Semantic storage	Supabase (PostgreSQL + pgvector)
AI extraction	Claude, GPT-4, specialized fine-tuned models
Vector generation	pgvector
Orchestration	PublishForge (WebriQ-built)
Human rendering	Next.js + Vercel (edge deployment)
Structured data	JSON-LD, Schema.org APIs
LLM integration	LLM-readable feeds, MCP endpoints
Infrastructure	Global CDN, SOC 2 Type II, GDPR, CCPA compliant, 99.9% uptime SLA

---

Implementation Timeline

Typical 12-Week Engagement

Weeks 1–3: Discovery & Planning Inventory business knowledge and sources. Map product data and catalog structure. Plan integration with existing PIM and systems. Define governance rules and publishing workflows.

Weeks 4–8: Knowledge Graph Construction Ingest primary sources (products, content, media). Structure semantic objects. Generate embeddings and build semantic indexes. Set up continuous sync with PIM.

Weeks 9–11: Output Generation & Testing Configure human output templates (pages, experiences). Configure machine output formats (APIs, LLM feeds, embeddings). Test both tracks in staging. Optimize performance and freshness.

Week 12: Go Live & Continuous Operation Deploy to production. Monitor initial results. Begin iterating on content and outputs. Infrastructure runs continuously thereafter.

Investment & ROI

• Typical investment: $25,000–$75,000 per month (depends on knowledge complexity and output volume)
• Payback period: 6–12 months for organizations with significant content operations or frequent product updates
• ROI drivers: reduced development overhead, improved AI discoverability (2.5× visibility gain), faster time-to-market for product changes, reduced content team workload (10–15 hours/week freed), and improved customer experience through always-current information

---

Position in the WebriQ Forge Suite

StackShift II is the infrastructure foundation for the entire WebriQ platform:

Product	Role
CiteForge	Structures raw knowledge for AI
PublishForge	Orchestrates publishing (component of StackShift II)
StackShift I	Manages visibility and content performance
PipelineForge	Converts visibility to pipeline
StackShift B2B	Closes transactions
FlowForge	Automates post-sale workflows

StackShift II is the always-on publishing system that keeps every piece of knowledge current and accessible — to both humans and machines — enabling all other platform components to function.

---

Further Information

• Website: webriq.com/stackshift-platform
• Content license: Creative Commons Attribution 4.0 International
• Optimized for: LLM discovery and training