Question 1

What is RCTDB and how is it different from a vector database?

Accepted Answer

RCTDB is an 8-dimensional AI memory schema designed specifically for constitutional AI workloads. A vector database (Pinecone, Weaviate) stores embeddings for semantic similarity search. RCTDB stores AI decision provenance โ€” query hashes, FDIA scores, model chains, consensus results, subject UUIDs, and timestamps. RCTDB enables PDPA right-to-erasure and Section 33 audit trails; vector databases do not.

Question 2

Can RCTDB do semantic search like Pinecone?

Accepted Answer

Yes. RCTDB includes semantic similarity search via the Delta Engine (similarity threshold 0.95). However, its primary purpose is not retrieval โ€” it is memory with governance. The 8-dimensional schema adds provenance, FDIA scoring, and consent management on top of semantic lookup.

Question 3

How does RCTDB achieve PDPA compliance that vector databases cannot?

Accepted Answer

RCTDB stores each record with a subject_uuid field. When a data subject requests erasure (PDPA Section 34), the UUID is tombstoned โ€” all future queries that would retrieve data for that subject receive nothing. Vector databases store embedding vectors with no native concept of data subject identity or erasure.

Question 4

What is the 8-dimensional RCTDB schema?

Accepted Answer

RCTDB stores 8 dimensions per record: (1) query_hash โ€” semantic fingerprint, (2) fdia_scores โ€” D/I/A/F values, (3) subject_uuid โ€” PDPA-compliant subject reference, (4) model_chain โ€” which LLMs were used, (5) consensus_result โ€” SignedAI agreement level, (6) delta_chain โ€” incremental state for 74% compression, (7) timestamp โ€” immutable creation time, (8) provenance โ€” source documentation and access log.

Question 5

What is Delta Engine compression and how does it save 74% storage?

Accepted Answer

The Delta Engine stores only what changed between consecutive query states rather than full snapshots. For stable enterprise query patterns (the majority of production workloads), this produces 74% average lossless compression versus always storing full state. Full state is reconstructed in sub-1ms when needed.

Feature	Pinecone	Weaviate	RCTDB
Semantic similarity search
PDPA subject UUID field (native)
Right-to-erasure (UUID tombstone)
Audit trail per query (automatic)
FDIA score storage (D/I/A/F)
Model chain provenance
SignedAI consensus tracking
Delta compression (74% lossless)
PDPA Section 33 evidence (auto)

Multi-tenant data isolation
Graph/relationship traversal
Yes Partial No

Scenario	Pinecone	Weaviate	RCTDB	Reason
Product recommendation engine	โ…	โ…	โ ๏ธ	Standard semantic retrieval โ€” use Pinecone or Weaviate. RCTDB is overkill unless PDPA compliance is required.
Enterprise AI with PDPA compliance (Thailand)	โ	โ	โ…	RCTDB is the only option with native subject_uuid, UUID tombstone erasure, and Section 33 provenance trail.
Multi-LLM agentic workflow memory	โ	โ	โ…	RCTDB stores model_chain and consensus_result โ€” enabling full reproducibility of multi-LLM decision chains.
Knowledge base for RAG system	โ…	โ…	โ…	All three work. RCTDB adds FDIA scoring and provenance to each retrieval, enabling constitutional AI compliance.
AI decision audit (regulatory requirement)	โ	โ	โ…	Only RCTDB automatically generates the audit trail required by financial, healthcare, and legal AI regulations.

RCTDB vs Vector Databases

Pinecone / Weaviate (Vector DB)

RCTDB (AI Memory Schema)

When to Use Both

Feature Comparison Matrix

When to Use Which

Explore the RCTDB Architecture