Skip to content

Implement an Azure AI Search Solution

This section of the Microsoft AI-102: Designing and Implementing a Microsoft Azure AI Solution exam covers building and managing Azure AI Search solutions. Below are study notes for each sub-topic, with links to Microsoft documentation, exam tips, and key facts

Provision an Azure AI Search Resource, Create an Index, and Define a Skillset

📖 Docs: Azure AI Search overview

Overview

  • Azure AI Search provides enterprise-grade search across structured and unstructured data
  • Workflow:
    • Provision an Azure AI Search resource
    • Define indexes (schema of searchable fields)
    • Create skillsets (AI enrichment such as OCR, entity recognition)

Key Points

  • Index schema fields can be:
    • searchable
    • filterable
    • sortable
    • facetable
  • Skillsets enrich content during indexing
  • Both prebuilt and custom skills supported

Exam Tip

Remember: Index = schema, Skillset = enrichment pipeline

Create Data Sources and Indexers

📖 Docs: Data sources and indexers

Overview

  • Data source: defines where documents come from (Blob Storage, Cosmos DB, SQL DB, etc.)
  • Indexer: pulls data from data source and pushes to index

Key Points

  • Indexers can run on schedule or be triggered manually
  • Apply skillsets during ingestion
  • Supports incremental indexing

Use Case

An indexer that ingests PDFs from Blob Storage and applies OCR skill

Implement Custom Skills and Include Them in a Skillset

📖 Docs: Custom skills in AI Search

Overview

  • Custom skills extend enrichment pipeline with user-defined logic
  • Hosted as Azure Functions, APIs, or Logic Apps

Key Points

  • Must conform to the skill interface (JSON in/out)
  • Fill gaps where prebuilt skills are insufficient
  • Example: custom ML model for sentiment analysis

Exam Tip

Scenario mentions custom logic enrichment → answer: custom skill

Create and Run an Indexer

📖 Docs: Create an indexer in Azure AI Search

Overview

  • Indexer orchestrates pipeline: data source → skillset → index
  • Execution can be:
    • Scheduled
    • On-demand

Key Points

  • Monitor status via portal or REST API
  • Error logs stored for failed documents
  • Supports incremental refresh

Best Practices

Configure retries and alerting for indexer failures

Query an Index (Syntax, Sorting, Filtering, Wildcards)

📖 Docs: Query index

Overview

  • Indexes are queried using REST API or SDK
  • Features:
    • Keyword search
    • Filters ($filter)
    • Sorting ($orderby)
    • Faceting (categories, counts)
    • Wildcards (*, ?)

Key Points

  • Supports Lucene syntax
  • Semantic ranking available
  • Vector search supported

Query Example

$filter=category eq 'Books' and price lt 20

Manage Knowledge Store Projections

📖 Docs: Knowledge store in Azure AI Search

Overview

  • Knowledge Store saves enriched documents for downstream use
  • Projection types:
    • File projections (images, documents)
    • Object projections (JSON objects)
    • Table projections (structured data)

Key Points

  • Projections stored in Blob Storage, Table Storage, or Cosmos DB
  • Enables analytics outside of search
  • Configured as part of skillset

Exam Tip

Scenario mentions storing enriched data → Knowledge Store

Implement Semantic and Vector Store Solutions

📖 Docs: Semantic search | Vector search

Overview

  • Semantic search = improves ranking by understanding meaning
  • Vector search = retrieves based on embeddings similarity

Key Points

  • Embeddings generated by models like Azure OpenAI
  • Hybrid search = keyword + vector search
  • Semantic config must be enabled in index

Use Case

RAG pipeline combining Azure AI Search + Azure OpenAI embeddings

Quick‑fire revision sheet

  • 📌 Index = schema, Skillset = enrichment, Indexer = pipeline
  • 📌 Data sources: Blob, Cosmos DB, SQL, etc.
  • 📌 Custom skills extend enrichment with APIs/Functions
  • 📌 Queries: filters, sorting, wildcards, facets
  • 📌 Knowledge Store saves enriched projections (file, object, table)
  • 📌 Semantic search = NLU ranking
  • 📌 Vector search = embeddings similarity
  • 📌 Hybrid search = semantic + vector + keyword