7.1. Implement Azure AI Search

💡 First Principle: Documents are opaque to computers—a PDF is just bytes, a Word doc is XML, an image is pixels. Azure AI Search transforms unstructured content into searchable structure through a pipeline: ingest → crack open → enrich with AI → index → query. The skillset is the key innovation: it's where AI "reads" your documents and creates searchable fields that didn't exist in the original content (like extracting people's names from contracts or locations from news articles).

What breaks without proper search configuration:

• Without an indexer schedule, content changes aren't reflected in search results (stale index)
• Without a skillset, PDFs remain opaque blobs you can't search inside
• Without proper field attributes (searchable, filterable, facetable), queries silently return no results
• Without semantic ranking, keyword searches miss conceptually related content ("car" won't find "automobile")

Imagine 10,000 contracts in a blob container. Users ask "Show me contracts expiring in Q3 with liability clauses over $1M." No single field in the original documents contains that information. With Search + AI enrichment, the skillset extracted expiration dates, identified clause types, and parsed dollar amounts—making them queryable. That's the power of knowledge mining.

This section covers the core pipeline, skillsets for AI enrichment, and knowledge store projections.