liveBook · Manning

Vector search profile

As explained in Enterprise RAG

A vector search profile is a search configuration that specifies how vector fields are indexed and searched.

A vector search profile is an Azure AI Search configuration that defines how a vector field is searched. Configure it with Exhaustive KNN and cosine similarity when a small dataset favors predictable search without HNSW’s additional memory structures. section 3.1

Pair the profile with a vector field whose dimensions match the generated embeddings. The demonstrated setup uses 1,536 dimensions, disables retrieval of the large vector array, and keeps ordinary fields available for filtering, sorting, and text retrieval. section 3.1

Apply the profile as part of an index design workflow:

  • Define fields for searchable text, structured filters, dates, and the vector representation. section 3.1
  • Set vector dimensions to match the embedding model and choose cosine similarity for direction-based semantic comparison. section 3.1
  • Choose Exhaustive KNN for smaller datasets; reserve HNSW’s faster large-scale search approach for cases that justify its additional index memory. section 3.1
  • Validate settings before saving because Azure AI Search does not allow later index edits after a configuration mistake. section 3.1

The profile supports a broader retrieval pipeline: transform SQL records into concise searchable text, generate embeddings, upload both representations, and combine vector similarity with text or semantic search. Keep the index synchronized by updating modified records and removing deleted ones. section 3.4

Figure 3.2 A correctly set-up Azure AI Search Index. Make sure to set the Retrievable, Filterable, Sortable, Facetable, and Searchable columns exactly as shown in this image.

Use the index configuration figure to verify retrievable, filterable, sortable, facetable, and searchable settings before creating the index. section 3.1

How Enterprise RAG covers it

How deep it goes
Explains dimensions, cosine similarity, Exhaustive KNN, HNSW tradeoffs, field flags, and index immutability.
Examples
Uses a 1,536-dimensional OpenAI embedding field with nonretrievable vectors, cosine similarity, and Exhaustive KNN for a small dataset.
What it assumes
Assumes embeddings and SQL records, while specifying vector dimensions, field capabilities, similarity, and search algorithms.
What you can do after
Operate/configure an Azure AI Search index for vector and hybrid retrieval while accounting for memory, field settings, and synchronization.
How it connects
Connects vector fields with embeddings, cosine similarity, Exhaustive KNN, HNSW, hybrid search, semantic search, and SQL records.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Enterprise RAG. The section links point to the parts of the book it draws on.