MongoDB Design Patterns: Advanced Strategies for AI Applications 2026 (Part 2 of 2)

				
					// Application code that handles multiple versions
async function getDocumentEmbedding(docId, preferredVersion = 'latest') {
  const doc = await db.documents.findOne({ "_id": docId })
  
  if (!doc) {
    throw new Error("Document not found")
  }
  
  // Handle different schema versions
  switch(doc.schema_version) {
    case 1:
      return {
        vector: doc.embedding_v1,
        dimensions: 768,
        model: "bert-base",
        needsMigration: true  // Flag for background migration
      }
    
    case 2:
      // Prefer V2 embedding, fallback to V1
      return {
        vector: doc.embedding_v2 || doc.embedding_v1,
        dimensions: doc.embedding_v2 ? 1536 : 768,
        model: doc.embedding_v2 ? "text-embedding-3" : "bert-base",
        needsMigration: !doc.embedding_v2  // Flag if V2 missing
      }
    
    case 3:
      // V3 has multiple embeddings, return requested one
      const embeddingKey = preferredVersion === 'text' ? 'text' : 'combined'
      return {
        vector: doc.embeddings[embeddingKey].vector,
        dimensions: doc.embeddings[embeddingKey].vector.length,
        model: doc.embeddings[embeddingKey].model,
        needsMigration: false
      }
    
    default:
      throw new Error(`Unknown schema version: ${doc.schema_version}`)
  }
}

// Background migration worker (gradual, non-blocking)
async function migrateDocumentsV1toV2(batchSize = 100) {
  const cursor = db.documents.find({
    "schema_version": 1,
    "migrated_from_v1": { $ne: true }
  }).limit(batchSize)
  
  for await (const doc of cursor) {
    try {
      // Generate new V2 embedding
      const newEmbedding = await openai.embeddings.create({
        model: "text-embedding-3-small",
        input: doc.text
      })
      
      // Extract metadata with AI
      const metadata = await extractMetadata(doc.text)
      
      // Update document to V2
      await db.documents.updateOne(
        { "_id": doc._id },
        {
          $set: {
            "schema_version": 2,
            "embedding_v2": newEmbedding.data[0].embedding,
            "metadata": metadata,
            "migrated_from_v1": true,
            "migration_date": new Date()
          }
        }
      )
      
      console.log(`Migrated document ${doc._id} to V2`)
      
    } catch (error) {
      console.error(`Migration failed for ${doc._id}:`, error)
      // Continue with next document
    }
  }
}

// Run migration in background (e.g., via cron job)
setInterval(() => {
  migrateDocumentsV1toV2(100)  // Migrate 100 docs every 5 minutes
}, 5 * 60 * 1000)