PROTOTYPE 14:42 UTC
Embeddings
Vector embeddings: model, dimensions, namespacing, query latency.
Total vectors
51,238
fleet-wide + 5 ships
Embedding model
text-embedding-3-small
OpenAI
Dimensions
1,536
per vector
Avg query latency
12ms
p95 28ms
Vectors by namespace
isolated per sailing · fleet-wide for shared knowledge
NamespaceVectorsLast upsertedQuery latency p50
fleet-wide (universal sailing)18,43213:1211ms
SLG-0427 Caribbean7,12814:3512ms
SLG-0428 Mediterranean6,89114:1813ms
SLG-0429 Alaska5,74214:0612ms
SLG-0430 Transatlantic6,41213:4811ms
SLG-0431 Bahamas6,63314:2213ms
Embedding pipeline

Documents are chunked at sentence-aware boundaries (±500 tokens), embedded via OpenAI text-embedding-3-small (1,536 dimensions), and upserted to Pinecone with rich metadata: source, date, content_type, and sailing_id. Queries run with top_k=8 and a 0.75 score threshold. Per-tenant namespace isolation — no cross-sailing inference.

Action confirmed.
PROTOTYPE · synthetic data · all guest records and entities are illustrative