Add AI-enriched fields

src/Elastic.Documentation.Configuration/DocumentationEndpoints.cs

@@ -45,4 +45,10 @@ public class ElasticsearchEndpoint
 	public int? BootstrapTimeout { get; set; }
 	public bool NoSemantic { get; set; }
 	public bool ForceReindex { get; set; }
+
+	/// <summary>
+	/// Enable AI enrichment of documents using LLM-generated metadata.
+	/// When enabled, documents are enriched with summaries, search queries, and questions.
+	/// </summary>
+	public bool EnableAiEnrichment { get; set; }
 }

src/Elastic.Documentation/Search/DocumentationDocument.cs

@@ -83,4 +83,49 @@ public record DocumentationDocument
 	[JsonPropertyName("hidden")]
 	[JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingDefault)]
 	public bool Hidden { get; set; }
+
+	// AI Enrichment fields - populated by DocumentEnrichmentService
+
+	/// <summary>
+	/// Content-addressable hash of title + body for AI enrichment cache lookup.
+	/// Used by the enrich processor to join AI-generated fields at index time.
+	/// </summary>
+	[JsonPropertyName("content_hash")]
+	[JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
+	public string? ContentHash { get; set; }
+
+	/// <summary>
+	/// 3-5 sentences dense with technical entities, API names, and core functionality for vector matching.
+	/// </summary>
+	[JsonPropertyName("ai_rag_optimized_summary")]
+	[JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
+	public string? AiRagOptimizedSummary { get; set; }
+
+	/// <summary>
+	/// Exactly 5-10 words for a UI tooltip.
+	/// </summary>
+	[JsonPropertyName("ai_short_summary")]
+	[JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
+	public string? AiShortSummary { get; set; }
+
+	/// <summary>
+	/// A 3-8 word keyword string representing a high-intent user search for this doc.
+	/// </summary>
+	[JsonPropertyName("ai_search_query")]
+	[JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
+	public string? AiSearchQuery { get; set; }
+
+	/// <summary>
+	/// Array of 3-5 specific questions answered by this document.
+	/// </summary>
+	[JsonPropertyName("ai_questions")]
+	[JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
+	public string[]? AiQuestions { get; set; }
+
+	/// <summary>
+	/// Array of 2-4 specific use cases this doc helps with.
+	/// </summary>
+	[JsonPropertyName("ai_use_cases")]
+	[JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
+	public string[]? AiUseCases { get; set; }
 }

src/Elastic.Markdown/Exporters/Elasticsearch/ElasticsearchIngestChannel.Mapping.cs

@@ -11,13 +11,15 @@ public abstract partial class ElasticsearchIngestChannel<TChannelOptions, TChann
 	where TChannelOptions : CatalogIndexChannelOptionsBase<DocumentationDocument>
 	where TChannel : CatalogIndexChannel<DocumentationDocument, TChannelOptions>
 {
-	protected static string CreateMappingSetting(string synonymSetName, string[] synonyms)
+	protected static string CreateMappingSetting(string synonymSetName, string[] synonyms, string? defaultPipeline = null)
 	{
 		var indexTimeSynonyms = $"[{string.Join(",", synonyms.Select(r => $"\"{r}\""))}]";
+		var pipelineSetting = defaultPipeline is not null ? $"\"default_pipeline\": \"{defaultPipeline}\"," : "";
 		// language=json
 		return
 			$$$"""
 			{
+				{{{pipelineSetting}}}
 				"analysis": {
 				  "normalizer": {
 					"keyword_normalizer": {
@@ -156,6 +158,7 @@ protected static string CreateMapping(string? inferenceId) =>
 		        }
 		      },
 		      "hash" : { "type" : "keyword" },
+		      "content_hash" : { "type" : "keyword" },
 		      "search_title": {
 		        "type": "text",
 		        "analyzer": "synonyms_fixed_analyzer",
@@ -201,6 +204,32 @@ protected static string CreateMapping(string? inferenceId) =>
 		        "fields" : {
 		          {{(!string.IsNullOrWhiteSpace(inferenceId) ? $"\"semantic_text\": {{{InferenceMapping(inferenceId)}}}" : "")}}
 		        }
+		      },
+		      "ai_rag_optimized_summary": {
+		        "type": "text",
+		        "analyzer": "synonyms_fixed_analyzer",
+		        "search_analyzer": "synonyms_analyzer",
+		        "fields": {
+		          {{(!string.IsNullOrWhiteSpace(inferenceId) ? $"\"semantic_text\": {{{InferenceMapping(inferenceId)}}}" : "")}}
+		        }
+		      },
+		      "ai_short_summary": {
+		        "type": "text"
+		      },
+		      "ai_search_query": {
+		        "type": "keyword"
+		      },
+		      "ai_questions": {
+		        "type": "text",
+		        "fields": {
+		          {{(!string.IsNullOrWhiteSpace(inferenceId) ? $"\"semantic_text\": {{{InferenceMapping(inferenceId)}}}" : "")}}
+		        }
+		      },
+		      "ai_use_cases": {
+		        "type": "text",
+		        "fields": {
+		          {{(!string.IsNullOrWhiteSpace(inferenceId) ? $"\"semantic_text\": {{{InferenceMapping(inferenceId)}}}" : "")}}
+		        }
 		      }
 		    }
 		  }

src/Elastic.Markdown/Exporters/Elasticsearch/ElasticsearchIngestChannel.cs

@@ -21,7 +21,8 @@ public class ElasticsearchLexicalIngestChannel(
 	ElasticsearchEndpoint endpoint,
 	string indexNamespace,
 	DistributedTransport transport,
-	string[] indexTimeSynonyms
+	string[] indexTimeSynonyms,
+	string? defaultPipeline = null
 )
 	: ElasticsearchIngestChannel<CatalogIndexChannelOptions<DocumentationDocument>, CatalogIndexChannel<DocumentationDocument>>
 	(logFactory, collector, endpoint, transport, o => new(o), t => new(t)
@@ -34,7 +35,7 @@ string[] indexTimeSynonyms
 				{ "batch_index_date", d.BatchIndexDate.ToString("o") }
 			}),
 		GetMapping = () => CreateMapping(null),
-		GetMappingSettings = () => CreateMappingSetting($"docs-{indexNamespace}", indexTimeSynonyms),
+		GetMappingSettings = () => CreateMappingSetting($"docs-{indexNamespace}", indexTimeSynonyms, defaultPipeline),
 		IndexFormat =
 			$"{endpoint.IndexNamePrefix.Replace("semantic", "lexical").ToLowerInvariant()}-{indexNamespace.ToLowerInvariant()}-{{0:yyyy.MM.dd.HHmmss}}",
 		ActiveSearchAlias = $"{endpoint.IndexNamePrefix.Replace("semantic", "lexical").ToLowerInvariant()}-{indexNamespace.ToLowerInvariant()}"
@@ -46,14 +47,15 @@ public class ElasticsearchSemanticIngestChannel(
 	ElasticsearchEndpoint endpoint,
 	string indexNamespace,
 	DistributedTransport transport,
-	string[] indexTimeSynonyms
+	string[] indexTimeSynonyms,
+	string? defaultPipeline = null
 )
 	: ElasticsearchIngestChannel<SemanticIndexChannelOptions<DocumentationDocument>, SemanticIndexChannel<DocumentationDocument>>
 	(logFactory, collector, endpoint, transport, o => new(o), t => new(t)
 	{
 		BulkOperationIdLookup = d => d.Url,
 		GetMapping = (inferenceId, _) => CreateMapping(inferenceId),
-		GetMappingSettings = (_, _) => CreateMappingSetting($"docs-{indexNamespace}", indexTimeSynonyms),
+		GetMappingSettings = (_, _) => CreateMappingSetting($"docs-{indexNamespace}", indexTimeSynonyms, defaultPipeline),
 		IndexFormat = $"{endpoint.IndexNamePrefix.ToLowerInvariant()}-{indexNamespace.ToLowerInvariant()}-{{0:yyyy.MM.dd.HHmmss}}",
 		ActiveSearchAlias = $"{endpoint.IndexNamePrefix}-{indexNamespace.ToLowerInvariant()}",
 		IndexNumThreads = endpoint.IndexNumThreads,

src/Elastic.Markdown/Exporters/Elasticsearch/ElasticsearchMarkdownExporter.Export.cs

@@ -8,6 +8,7 @@
 using Elastic.Documentation.Navigation;
 using Elastic.Documentation.Search;
 using Elastic.Ingest.Elasticsearch.Indices;
+using Elastic.Markdown.Exporters.Elasticsearch.Enrichment;
 using Elastic.Markdown.Helpers;
 using Markdig.Syntax;
 using Microsoft.Extensions.Logging;
@@ -131,6 +132,13 @@ public async ValueTask<bool> ExportAsync(MarkdownExportFileContext fileContext,
 		};
 
 		CommonEnrichments(doc, currentNavigation);
+
+		// AI Enrichment - hybrid approach:
+		// - Cache hits: enrich processor applies fields at index time
+		// - Cache misses: apply fields inline before indexing
+		doc.ContentHash = ContentHashGenerator.Generate(doc.Title, doc.StrippedBody ?? string.Empty);
+		await TryEnrichDocumentAsync(doc, ctx);
+
 		AssignDocumentMetadata(doc);
 
 		if (_indexStrategy == IngestStrategy.Multiplex)
@@ -166,6 +174,11 @@ public async ValueTask<bool> FinishExportAsync(IDirectoryInfo outputFolder, Canc
 			doc.Abstract = @abstract;
 			doc.Headings = headings;
 			CommonEnrichments(doc, null);
+
+			// AI Enrichment - hybrid approach
+			doc.ContentHash = ContentHashGenerator.Generate(doc.Title, doc.StrippedBody ?? string.Empty);
+			await TryEnrichDocumentAsync(doc, ctx);
+
 			AssignDocumentMetadata(doc);
 
 			// Write to channels following the multiplex or reindex strategy
@@ -191,4 +204,51 @@ public async ValueTask<bool> FinishExportAsync(IDirectoryInfo outputFolder, Canc
 		return true;
 	}
 
+	/// <summary>
+	/// Hybrid AI enrichment: cache hits rely on enrich processor, cache misses apply fields inline.
+	/// </summary>
+	private async ValueTask TryEnrichDocumentAsync(DocumentationDocument doc, Cancel ctx)
+	{
+		if (_enrichmentCache is null || _llmClient is null || string.IsNullOrWhiteSpace(doc.ContentHash))
+			return;
+
+		// Check if enrichment exists in cache
+		if (_enrichmentCache.Exists(doc.ContentHash))
+		{
+			// Cache hit - enrich processor will apply fields at index time
+			_ = Interlocked.Increment(ref _cacheHitCount);
+			return;
+		}
+
+		// Check if we've hit the limit for new enrichments
+		var current = Interlocked.Increment(ref _newEnrichmentCount);
+		if (current > _enrichmentOptions.MaxNewEnrichmentsPerRun)
+		{
+			_ = Interlocked.Decrement(ref _newEnrichmentCount);
+			return;
+		}
+
+		// Cache miss - generate enrichment inline and apply directly
+		try
+		{
+			var enrichment = await _llmClient.EnrichAsync(doc.Title, doc.StrippedBody ?? string.Empty, ctx);
+			if (enrichment is not { HasData: true })
+				return;
+
+			// Store in cache for future runs
+			await _enrichmentCache.StoreAsync(doc.ContentHash, enrichment, _enrichmentOptions.PromptVersion, ctx);
+
+			// Apply fields directly (enrich processor won't have this entry yet)
+			doc.AiRagOptimizedSummary = enrichment.RagOptimizedSummary;
+			doc.AiShortSummary = enrichment.ShortSummary;
+			doc.AiSearchQuery = enrichment.SearchQuery;
+			doc.AiQuestions = enrichment.Questions;
+			doc.AiUseCases = enrichment.UseCases;
+		}
+		catch (Exception ex) when (ex is not OperationCanceledException)
+		{
+			_logger.LogWarning(ex, "Failed to enrich document {Url}", doc.Url);
+			_ = Interlocked.Decrement(ref _newEnrichmentCount);
+		}
+	}
 }