[Bastion-RAG 4 – Anchor] Embedding Model Bias Verification
As Retrieval-Augmented Generation (RAG) architectures pairing Large Language Models (LLMs) with dense vector retrieval establish...
[Bastion-RAG 4 – Anchor] Embedding Noise Injection
I consider myself a business-friendly security engineer. Granted, there are sectors—such as those dealing with...
[Bastion-RAG 3 – Navigator] Logical Partitioning
The architectural modules spanning from logical partitioning to embedding model bias validation are highly theoretical....
[Bastion-RAG 3 – Navigator] Hybrid Reranking
Technically speaking, “hybrid reranking” is not an inherently complex concept, and to be honest, it...
[Bastion-RAG 2 – Vault] – Deterministic De-identification
Looking at recent security incidents, I find myself questioning whether this type of de-identification is...
[Bastion-RAG 2 – Vault] Multi-Tenancy
As software engineers, confronting new architectural paradigms and implementing them into live production code invariably...
[Bastion-RAG 1 – Sentinel] – Metadata Filtering
When designing “Metadata Filtering,” defining its conceptual boundaries was initially challenging. Unlike “Prompt Injection,” which...
[Bastion-RAG 1 – Sentinel] Prompt Injection Defense
This section marks the beginning of the implementation phase. While the complete architecture design ([Bastion-RAG...
[Bastion-RAG 0] Get help from AI (Architecture Design)
This section marks the beginning of the implementation phase. For this reason, the comprehensive architecture...
10 Technical Reports for Token Consumption and Cost Optimization in Claude Code and Gemini CLI Environments
Terminal-based AI agents like Claude Code and Gemini Cli drastically boost our productivity by analyzing...