@memberjunction/ai-vector-sync

A robust MemberJunction package for synchronizing entities with vector databases by transforming entity records into vector representations using embedding models.

Overview

The @memberjunction/ai-vector-sync package provides a comprehensive solution for:

• Converting MemberJunction entities into vector embeddings
• Storing embeddings in vector databases (currently supports Pinecone)
• Managing the synchronization lifecycle between entities and their vector representations
• Supporting batch processing for large datasets
• Providing template-based document generation for vectorization

Installation

npm install @memberjunction/ai-vector-sync

Prerequisites

Before using this package, ensure you have:

1. SQL Database with MemberJunction Framework
   A properly configured SQL database with the MemberJunction framework installed.

2. API Keys

   • Embedding model API key (supports OpenAI, Mistral, etc.)
   • Vector database API key (currently supports Pinecone)

3. Entity Configuration

   • Entity Document record defined in MemberJunction
   • Associated template for specifying which entity properties to vectorize

Core Features

Entity Vectorization

Transform entity records into high-dimensional vectors that capture the semantic meaning of the data.

Batch Processing

Efficiently handle large datasets with configurable batch sizes for:

• Record fetching
• Vectorization
• Database upsertion

Template-Based Processing

Use MemberJunction templates to define which entity fields and relationships to include in vectorization.

Vector Database Integration

Seamlessly integrate with vector databases through the MemberJunction AI infrastructure.

Usage

Basic Entity Vectorization

import { EntityVectorSyncer } from '@memberjunction/ai-vector-sync';
import { UserInfo } from '@memberjunction/core';

// Initialize the syncer
const syncer = new EntityVectorSyncer();

// Configure the syncer (required before first use)
await syncer.Config(false, contextUser);

// Vectorize an entity
const params = {
  entityID: 'your-entity-id',
  entityDocumentID: 'your-entity-document-id',
  listBatchCount: 50,        // Optional: records per batch (default: 50)
  VectorizeBatchCount: 50,   // Optional: vectorization batch size (default: 50)
  UpsertBatchCount: 50,      // Optional: upsert batch size (default: 50)
  StartingOffset: 0          // Optional: skip records for resuming
};

// Start vectorization (runs asynchronously)
syncer.VectorizeEntity(params, contextUser);

Vectorizing a Specific List

// Vectorize only records within a specific list
const params = {
  entityID: 'your-entity-id',
  entityDocumentID: 'your-entity-document-id',
  listID: 'your-list-id',    // Only vectorize records in this list
  listBatchCount: 100
};

await syncer.VectorizeEntity(params, contextUser);

Working with Entity Documents

// Get entity document by ID
const entityDoc = await syncer.GetEntityDocument('document-id');

// Get entity document by name
const entityDoc = await syncer.GetEntityDocumentByName('Document Name', contextUser);

// Get all active entity documents
const activeDocs = await syncer.GetActiveEntityDocuments();

// Get active documents for specific entities
const specificDocs = await syncer.GetActiveEntityDocuments(['Entity1', 'Entity2']);

Creating Default Entity Documents

import { VectorDatabaseEntity, AIModelEntity } from '@memberjunction/core-entities';

// Create a default entity document when one doesn't exist
const entityDoc = await syncer.CreateDefaultEntityDocument(
  entityID,
  vectorDatabase,  // VectorDatabaseEntity instance
  aiModel         // AIModelEntity instance
);

API Reference

EntityVectorSyncer

The main class for entity vectorization operations.

Methods

Config(forceRefresh: boolean, contextUser?: UserInfo): Promise<void>

Configures the syncer and initializes required engines.

• forceRefresh: Force refresh of caches and engines
• contextUser: User context for operations

VectorizeEntity(params: VectorizeEntityParams, contextUser?: UserInfo): Promise<VectorizeEntityResponse>

Vectorizes entities based on provided parameters.

• params: Configuration for vectorization
• contextUser: Required user context

GetEntityDocument(entityDocumentID: string): Promise<EntityDocumentEntity | null>

Retrieves an entity document by ID.

GetEntityDocumentByName(entityDocumentName: string, contextUser?: UserInfo): Promise<EntityDocumentEntity | null>

Retrieves an entity document by name.

GetActiveEntityDocuments(entityNames?: string[]): Promise<EntityDocumentEntity[]>

Gets all active entity documents, optionally filtered by entity names.

CreateDefaultEntityDocument(entityID: string, vectorDatabase: VectorDatabaseEntity, aiModel: AIModelEntity): Promise<EntityDocumentEntity>

Creates a default entity document for the specified entity.

Types

VectorizeEntityParams

type VectorizeEntityParams = {
  entityID: string;              // Required: Entity to vectorize
  entityDocumentID?: string;     // Entity document configuration
  listID?: string;               // Optional: Specific list to vectorize
  listBatchCount?: number;       // Records per fetch batch (default: 50)
  VectorizeBatchCount?: number;  // Vectorization batch size (default: 50)
  UpsertBatchCount?: number;     // Database upsert batch size (default: 50)
  StartingOffset?: number;       // Skip records for resuming
  CurrentUser?: UserInfo;        // User context
  options?: any;                 // Additional options
}

EntitySyncConfig

type EntitySyncConfig = {
  EntityDocumentID: string;      // Entity document to use
  Interval: number;              // Sync interval in seconds
  RunViewParams: RunViewParams;  // View parameters for fetching records
  IncludeInSync: boolean;        // Include in sync process
  LastRunDate: string;           // Last sync timestamp
  VectorIndexID: number;         // Vector index ID
  VectorID: number;              // Vector database ID
}

Architecture

Process Flow

1. Entity Document Retrieval: Fetches configuration from Entity Document record
2. Model and Database Configuration: Sets up embedding model and vector database
3. Data Fetching: Retrieves entity records in batches
4. Vectorization: Transforms records using embedding model
5. Vector Upsertion: Stores vectors in database
6. EntityRecordDocument Creation: Creates tracking records

Worker Architecture

The package uses a multi-worker architecture for efficient processing:

• VectorizeTemplates Worker: Handles template-based text generation and embedding
• UpsertVectors Worker: Manages vector database operations
• EntityRecordDocument Worker: Tracks vector-entity relationships

Configuration

Environment Variables

Create a .env file with:

# Database Configuration
DB_HOST=your-database-host
DB_PORT=1433
DB_USERNAME=your-username
DB_PASSWORD=your-password
DB_DATABASE=your-database

# API Keys
OPENAI_API_KEY=your-openai-key
MISTRAL_API_KEY=your-mistral-key
PINECONE_API_KEY=your-pinecone-key
PINECONE_HOST=your-pinecone-host
PINECONE_DEFAULT_INDEX=your-default-index

# User Configuration
CURRENT_USER_EMAIL=[e-mail]

Performance Considerations

• Long-Running Processes: Vectorization can take hours for large datasets
• Batch Sizes: Adjust batch sizes based on your system resources
• Asynchronous Processing: Consider running vectorization in background processes
• Memory Usage: Monitor memory usage for large batch sizes

Integration with MemberJunction

This package integrates seamlessly with:

• @memberjunction/core: Core entity and metadata functionality
• @memberjunction/ai: AI model abstractions
• @memberjunction/ai-vectordb: Vector database abstractions
• @memberjunction/templates: Template processing engine

Error Handling

The package includes comprehensive error handling:

• Validation of entity documents and templates
• Graceful handling of API failures
• Detailed logging through MemberJunction's logging system

Best Practices

1. Start with Small Batches: Test with small batch sizes before processing large datasets
2. Monitor Progress: Use MemberJunction's logging to track vectorization progress
3. Handle Interruptions: Use StartingOffset to resume interrupted processes
4. Template Design: Design templates to include relevant fields for semantic search
5. Resource Management: Consider database and API rate limits when setting batch sizes

License

ISC - See LICENSE file for details

Author

MemberJunction.com