Build CustomDocument RAG Pipelines.

using LMKit.Model;
using LMKit.Retrieval;
using LMKit.Data.Storage;
using LMKit.TextGeneration;

// Load models
var chatModel = LM.LoadFromModelID("qwen3:4b");
var embedModel = LM.LoadFromModelID("embeddinggemma-300m");
var visionModel = LM.LoadFromModelID("qwen3-vl:4b");

// Configure vector store (optional - uses in-memory by default)
var vectorStore = new FileSystemVectorStore("./embeddings");

// Create DocumentRag with full configuration
var docRag = new DocumentRag(embedModel, vectorStore)
{
    ProcessingMode = PageProcessingMode.Auto,
    MaxChunkSize = 512,
    OcrEngine = new OcrEngine(),
    VisionParser = new VlmOcr(visionModel)
};

// Subscribe to progress events
docRag.Progress += (sender, e) =>
{
    if (e.Phase == DocumentImportPhase.PageProcessingStarted)
        Console.WriteLine($"Processing page {e.PageIndex + 1}/{e.TotalPages}");
};

// Import document with explicit ID for lifecycle management
var attachment = Attachment.FromFile("financial-report.pdf");
var metadata = new DocumentRag.DocumentMetadata(
    attachment,
    id: "fin-report-2024-q4",
    sourceUri: "https://internal.company.com/reports/q4-2024.pdf");

await docRag.ImportDocumentAsync(attachment, metadata, "financial-docs");

// Find relevant passages
var partitions = await docRag.FindMatchingPartitionsAsync(
    "What was the Q4 revenue?",
    topK: 5,
    minScore: 0.5f);

// Generate response with source attribution
var conversation = new SingleTurnConversation(chatModel);
var result = await docRag.QueryPartitionsAsync(
    "What was the Q4 revenue?",
    partitions,
    conversation);

Console.WriteLine(result.Response.Completion);
foreach (var reference in result.SourceReferences)
{
    Console.WriteLine($"  Source: {reference.Name}, Page {reference.PageNumber}");
    Console.WriteLine($"  Score: {reference.SimilarityScore:P1}");
}

// Later: Delete document by ID
await docRag.DeleteDocumentAsync("fin-report-2024-q4", "financial-docs");