SYSTEM Cited by 3 sources

Mosaic AI Vector Search (Databricks)

Mosaic AI Vector Search is Databricks' managed vector-search service over governed Delta tables. Vector embeddings produced upstream (radiomics, deep-model feature extractors, text embeddings, etc.) are stored as Delta table columns; Vector Search indexes them and serves similarity queries without moving the vectors out of Unity Catalog's governance domain.

Stub page. First wiki ingest naming Mosaic AI Vector Search; the ingested Databricks multimodal post uses it as the imaging- similarity modality tool inside the governed-Delta-tables-per- modality pattern.

Role in multimodal lakehouse architecture

"For imaging, the pattern is: (1) derive features / embeddings upstream (radiomics or deep model outputs), (2) store features as governed Delta tables (secured via Unity Catalog), and (3) use vector search for similarity queries (e.g., 'find similar phenotypes within glioblastoma')." (Source: sources/2026-04-22-databricks-multimodal-data-integration-production-architectures-for-healthcare-ai)

Key architectural properties:

• Vectors live in Delta tables — no separate vector store, no separate governance, no extra copy.
• Indexed under UC — the same ACL / tag / audit controls that apply to the underlying Delta table apply to the index.
• Similarity queries are the access path — "find similar phenotypes", "find similar cases", "find similar claims" shapes fit without exporting embeddings into a parallel system.

Why it matters for sysdesign

The "vectors-as-Delta-columns + UC-governed index" posture is the architectural move that keeps vector search out of the specialty-store-per- modality anti-pattern. Separate vector DBs (Pinecone, Weaviate, Milvus, Atlas Vector Search as a standalone tier) reintroduce the cross-store governance burden; Mosaic AI Vector Search sidesteps it by making vectors a first-class column type on governed tables.

Comparable systems on the wiki:

• systems/atlas-vector-search — MongoDB's embedding search product; independent store shape.
• systems/cloudflare-vectorize — Cloudflare's vector DB; also independent store.
• systems/s3-vectors — AWS's vectors-on-S3 offering; similar "don't make vectors a separate stack" philosophy.
• systems/pinterest-vector-db-service — internal vector-DB service at Pinterest.

Seen in

• sources/2026-05-22-databricks-how-world-bank-group-uses-databricks-to-eradicate-poverty-through-shared-knowledge — RAG-Agent-in-multi-Genie-fronted-by-agentic-router face. World Bank Group uses Mosaic AI Vector Search as the document-similarity substrate for the RAG agent inside its agentic-router-fronted multi-domain knowledge platform. Indexed corpus is "tens of millions" of project documents stored in UC Volumes; the RAG agent serves as a peer of the per-domain Genie sub-agents in the routing graph. Operational scale: 3M document downloads / month through the AI-powered layer, half from low- and middle-income countries. This is the wiki's first canonicalised case of Mosaic AI Vector Search inside the fan-out-and-decompose generalisation of multi-agent routing (vs the alternative-selection shape from Virtue Foundation). Pattern instance: patterns/intent-domain-decomposer-agentic-router. Caveat: embedding model, chunking strategy, indexing cadence, and per- query latency / concurrency not disclosed.

• sources/2026-05-20-databricks-virtue-foundation-medical-volunteers-72-countries — Vector-Search-Agent-in-multi-agent-supervisor face. Mosaic AI Vector Search is the substrate for the Vector Search Agent sub-agent inside Virtue Foundation's VF Agent prototype. In the four-sub-agent composition (Medical Specialty Extractor + Multi-Agent Supervisor

• Vector Search Agent + Genie Agent), the supervisor routes facility-discovery and search queries ("find me orthopedic volunteer opportunities in Ghana with X-ray equipment available") to the Vector Search Agent, while analytical / structured queries route to the Genie Agent. Canonicalises Mosaic AI Vector Search at the multi-agent supervisor routing altitude — distinct from the prior 2026-04-22 framing as imaging-similarity modality tool. The two faces share the vectors-as-Delta-columns + UC-governed index substrate but expose it differently: imaging-similarity to a research query workflow vs facility-discovery to a multi-agent NL-query system. See patterns/multi-agent-supervisor-routing.

• sources/2026-04-22-databricks-multimodal-data-integration-production-architectures-for-healthcare-ai — Databricks frames Mosaic AI Vector Search as the imaging- similarity access path inside its multimodal lakehouse; vectors land in Delta, are governed by UC, and are queried for similarity without leaving the lakehouse. Illustrative query shape: "find similar phenotypes within glioblastoma" (cohort-discovery + retrospective-comparison use case). First wiki ingest naming Mosaic AI Vector Search.