ai-korp-neural-archives

AI-KORP | Neural Asset Archives

Applied Research: Investigating automated design taxonomies using AWS cloud infrastructure for AEC and Branding.

1. Executive Summary

This repository serves as a technical log for a proof-of-concept study. The objective is to evaluate how cloud-native AI services can transition static architectural and branding archives into “Active Knowledge Bases.”

By utilizing a serverless pipeline, we move from manual filing to a system that “understands” architectural intent, material composition, and brand application.

2. Theoretical Framework: The AI-KORP Baskets

This study operationalizes the AI-KORP Framework, specifically focusing on the Intellectual (Knowledge) and Spatial (Context) baskets.

Input: Raw design assets (Renders, Brand Mockups, Site Photos).
Process: AWS-driven extraction of visual and contextual metadata.
Output: A searchable, tagged ecosystem for design synthesis.

3. The Technical Stack (AWS)

To ensure professional data sovereignty, the pipeline is built on a modular, private cloud environment:

Service	Role
Amazon S3	Object storage for high-resolution design assets.
AWS Lambda	Event-driven logic to trigger analysis on upload.
Amazon Rekognition	Computer vision for label detection (Materials, Objects, Styles).
Amazon Bedrock	LLM reasoning to synthesize raw tags into design summaries.

4. Visual Study & Analysis

Below are the sample inputs used to test the pipeline’s granularity:

Asset A: Brand Archetype

Testing text detection and graphic consistency. Brand Toolkit

Asset B: Spatial Context (Industrial Workshop)

Testing complex environment recognition (Gantries, Concrete, Workflow). Workshop Render

5. Compliance & Governance

This research is conducted with a “Security-First” mindset:

Data Privacy: All processing occurs within a private VPC.
Ethics: Aligned with the EU AI Act regarding transparent data processing and human-in-the-loop verification.

🔍 Live Audit: Machine Perception

I analyzed the primary workshop render using Amazon Rekognition to extract a spatial taxonomy.

Top High-Confidence Labels:

Architecture: 99.9%
Factory / Industrial Building: 99.9%
Person: 98.6% (Validating human scale)
Manufacturing: 72.8%

Raw Data Archive: Download the verified architectural taxonomy: Master Label List (JSON)

Note: This is a non-commercial research project exploring workflow optimization for the AEC and Brand Design sectors.

System Architecture

To ensure professional-grade security, the pipeline is built on a modular AWS environment. This “sandboxed” approach ensures that design assets remain private and are excluded from public AI training loops.

```mermaid graph TD %% Ingestion Layer A[Design Asset: Render/Photo] –>|Upload| B(Amazon S3: Research-Archive)

%% Logic Layer
B -->|S3 Event Trigger| C{AWS Lambda: Boto3 Script}

%% Analysis Layer
C -->|Analyze Image| D[Amazon Rekognition: Vision Engine]
D -->|Extract Labels| C

%% Reasoning Layer
C -->|Synthesize Context| E[Amazon Bedrock: Claude 3.5]
E -->|Generate Design Summary| C

%% Output Layer
C -->|Final Metadata| F[(Neural Archive Database)]
F --> G{Creative Basket}
F --> H{Spatial Basket}
F --> I{Intellectual Basket}

%% AWS Branding Colors
style B fill:#FF9900,stroke:#232F3E,color:#fff
style C fill:#FF9900,stroke:#232F3E,color:#fff
style D fill:#3F8624,stroke:#232F3E,color:#fff
style E fill:#D05C45,stroke:#232F3E,color:#fff