NGIE-AI Current Platform Architecture

Snapshot date: July 6, 2026
Detailed reference: docs/NGIE_AI_ARCHITECTURE_OVERVIEW.md

What NGIE-AI Is Today

NGIE-AI is now a deterministic Business Knowledge Operating System that can ingest ecosystem websites, convert them into structured knowledge, build relationships, generate business intelligence, make rule-based recommendations, and produce client-ready ecosystem assessments without using AI APIs yet.

Websites
   ↓
Extraction Pipeline
   ↓
Knowledge Objects
   ↓
Search + Review + Explorer
   ↓
Knowledge Graph
   ↓
Business Intelligence
   ↓
Decision Engine
   ↓
Ecosystem Assessment
   ↓
Future Client Products

The current implementation includes an Industry Solutions bridge between the Decision Engine and Ecosystem Assessment.

Folder Flow

sources/
   ↓
knowledge/
   ↓
knowledge/objects/
   ↓
knowledge/relationships/     (logical; relationships currently live in objects)
   ↓
knowledge/graph/
   ↓
reports/
   ↓
reports/client-assessments/

Current Platform Summary

Area	Current status
Website pages processed	137 across 6 websites
Metadata files	137
Canonical knowledge objects	137
Review	29 approved; 100 pending; 3 needs revision; 5 rejected
Review warnings	0 unresolved thin objects; 0 active duplicates
Graph	859 nodes; 2,958 edges; 1 component; no orphans or dangling edges
Business Intelligence	21.2% approved; 6 organizations; not ready for AI reasoning
Decision Engine	6 deterministic business profiles
Industry Solutions	6 blueprints; deployment disabled
Ecosystem Assessment	Operational and tested; JSON, Markdown, PDF placeholder
Workflow Engine	3 deterministic workflows; allow-list, dry-run, fail-fast, and audit records
Sample assessment	39/100; 17 approved supporting objects
ChatGPT preparation	Ready, but no official export is present

Component Status

#	Component	Main input	Main output	Status
1	Website Source Registry	sources/websites.yaml	Active source definitions	Complete
2	Website Knowledge Extraction	Registry and websites	knowledge/<site>/{raw,cleaned,pages,metadata,reports}	Operational
3	ChatGPT Export Preparation	Future conversations.json	Private import folders and readiness check	Prepared; awaiting export
4	Website Knowledge Object Generator	Page Markdown and metadata	knowledge/objects/*.json	Operational; 137 objects
5	Local Knowledge Search	Canonical objects	Ranked local results and optional index	Operational
6	Knowledge Review Workflow	Canonical objects	Review fields and status reports	Operational
7	Knowledge Explorer	Objects and ID map	Interactive terminal exploration	Operational
8	Knowledge Graph	Objects and relationships	Graph JSON/JSONL and metrics	Operational
9	Business Intelligence	Objects and graph	intelligence/* and BI report	Operational; gated
10	Decision Engine	Objects, graph, and BI	Six business profiles	Operational
11	Ecosystem Assessment	Client input and all downstream platform layers	Client assessment JSON, Markdown, and PDF	Operational and tested
12	Workflow Engine	Workflow registry and local platform dependencies	Audited workflow runs and step logs	Operational and tested

Component Connections

1. The registry selects active sources.
2. The crawler turns websites into raw HTML, cleaned HTML, Markdown, and metadata.
3. The object generator normalizes extracted pages into the canonical schema.
4. Search, review, and the Explorer expose and govern canonical objects.
5. The graph materializes object relationships.
6. BI summarizes quality, coverage, organizations, and graph signals.
7. The Decision Engine creates organization profiles and rule-based candidates.
8. Industry Solutions package approved evidence into industry blueprints.
9. The Ecosystem Assessment combines declared client inputs with approved evidence and produces client-facing artifacts.
10. The Workflow Engine makes refresh, validation, and assessment execution reproducible without adding autonomous behavior.

Architecture Observations Requiring Validation

• knowledge/objects/ contains 137 canonical objects.
• knowledge/classification-index.json and the persisted search index contain 322 derived documents. These must remain clearly separated from canonical counts.
• There is no physical knowledge/relationships/ directory. Relationship fields are stored in objects and graph edges.
• Website caps left 1,061 discovered URLs queued or deferred.
• Only 29 canonical objects are approved, so BI correctly reports that the repository is not ready for AI reasoning.
• Assessment recommendations are evidence-linked and deterministic, but the PDF is still a placeholder.
• The official private ChatGPT export has not been imported.

Recommended Next Build

1. Architecture validation report.
2. Test assessment using one sample business.
3. Assessment report template improvements.
4. Simple local dashboard.
5. AI reasoning layer after validation.

This report documents current state only. It does not introduce new runtime features.