WordPress Environment

• WordPress: 6.0+ (tested up to latest)
• PHP: 8.0+ (for modern syntax and performance)
• MySQL/MariaDB: 5.7+ / 10.3+ (for metadata storage)
• Server: Apache or Nginx with mod_rewrite
• Memory: 256MB PHP memory minimum (512MB recommended)

Vector Database (Choose One)

Embedding API

• OpenAI API: text-embedding-3-large or text-embedding-3-small
• Cost: ~$0.13 per 1M tokens (3-large) or $0.02 per 1M tokens (3-small)
• Dimensions: 3072 (3-large) or 1536 (3-small)
• Alternative: Cohere, Voyage AI, or open-source models

Development Tools

• Git: Version control
• Composer: PHP dependency management
• npm/yarn: For any JavaScript components
• WP-CLI: WordPress command-line interface (optional but helpful)
• Postman/Insomnia: API testing

1. Content Extraction Hooks

// Hook into post save
add_action('save_post', 'sl_process_content', 10, 3);
add_action('post_updated', 'sl_update_vector', 10, 3);
add_action('before_delete_post', 'sl_delete_vector', 10, 1);

function sl_process_content($post_id, $post, $update) {
    // Extract title, meta description, content
    // Generate embedding
    // Store in vector DB
}

2. Vector Database Connector

class Vector_DB {
    private $client;
    
    public function upsert($post_id, $vector, $metadata) {}
    public function search($vector, $top_k = 10, $filter = []) {}
    public function delete($post_id) {}
    public function get_stats() {}
}

3. Embedding Generator

class Embedding_Generator {
    private $api_key;
    private $model = 'text-embedding-3-large';
    
    public function generate($text) {
        // Call OpenAI API
        // Handle rate limiting
        // Return vector array
    }
    
    public function batch_generate($texts) {
        // Process multiple texts efficiently
    }
}

4. Recommendation Engine

class Recommender {
    public function get_recommendations($post_id, $options = []) {
        // 1. Get post vector from DB
        // 2. Query vector DB for similar
        // 3. Apply filters (practice area, location, etc.)
        // 4. Apply business rules
        // 5. Return ranked recommendations
    }
    
    private function apply_filters($results, $filters) {}
    private function apply_business_rules($results) {}
    private function rank_results($results) {}
}

OpenAI API Setup

1. Create account at platform.openai.com
2. Add payment method (required for API access)
3. Generate API key with embedding permissions
4. Set spending limits ($10-50/month typical for medium law firm)
5. Store API key in WordPress wp-config.php or environment variables

// In wp-config.php
define('OPENAI_API_KEY', 'sk-...');

Vector Database API Setup (Pinecone Example)

1. Sign up at pinecone.io
2. Create index with 3072 dimensions, cosine metric
3. Generate API key
4. Note your environment region
5. Store credentials securely

// In wp-config.php
define('PINECONE_API_KEY', 'xxx');
define('PINECONE_ENVIRONMENT', 'us-west1-gcp');
define('PINECONE_INDEX_NAME', 'semantic-linking');

1. Settings Page

• API credentials configuration
• Vector database selection & connection
• Embedding model selection
• Similarity threshold slider (0.0 – 1.0)
• Practice area taxonomy mapping
• Location filtering rules
• Business rules configuration
• Auto-linking enable/disable toggle

2. Dashboard Page

• Total posts embedded count
• Embedding API usage & cost
• Vector DB storage metrics
• Recent embedding activity log
• Recommendation acceptance rates
• Top performing content clusters
• System health status

3. Batch Processing Page

• Bulk embed all posts button
• Filter by post type, category, date range
• Progress bar with estimated completion
• Pause/resume functionality
• Error handling & retry logic
• Export/import embeddings (backup)

4. Post Editor Meta Box

• Show recommendations for current post
• Display similarity scores
• One-click insert link buttons
• Manual regenerate embedding button
• Practice area/location override
• Exclude from recommendations toggle

5. Analytics Page

• Most recommended posts
• Recommendation click-through rates
• Content cluster visualization
• Practice area distribution charts
• User engagement impact metrics
• ROI calculator based on traffic improvements

✅ Functional Testing

• New post auto-embeds on publish
• Updated post re-embeds correctly
• Deleted post removes vector from DB
• Recommendations appear in post editor
• Similarity scores calculate correctly
• Filters (practice area, location) work as expected
• One-click link insertion functions properly
• Batch processing completes successfully
• Cache invalidation works correctly

⚡ Performance Testing

• Single embedding generation < 2 seconds
• Vector search query < 500ms
• Recommendation display in post editor < 1 second
• Batch processing handles 1000+ posts without timeout
• Memory usage stays under PHP limits
• Database queries optimized (no N+1 issues)
• Caching reduces API calls by 80%+

🔒 Security Testing

• API keys not exposed in client-side code
• Nonces validated on all AJAX requests
• Capability checks enforce permissions
• Input sanitization prevents XSS
• SQL queries use prepared statements
• Rate limiting prevents API abuse
• Error messages don’t leak sensitive info

🎯 Accuracy Testing

• Semantically similar content scores high (>0.75)
• Unrelated content scores low (<0.50)
• Practice area filtering works correctly
• Geographic filtering prevents cross-state links
• Business rules applied consistently
• Manual review of 50 random recommendations confirms relevance

🔄 Compatibility Testing

• Works with Gutenberg editor
• Compatible with Classic Editor plugin
• No conflicts with Yoast SEO / Rank Math
• Works alongside caching plugins (WP Rocket, etc.)
• Compatible with common page builders (Elementor, etc.)
• Tested on PHP 8.0, 8.1, 8.2
• Works with WordPress 6.0, 6.1, 6.2+

☑️ Pre-Launch

• All tests passing (functional, performance, security)
• Production API keys configured
• Vector database production index created
• Backup strategy in place
• Rollback plan documented
• Monitoring and alerts configured
• Team training completed

☑️ Launch Day

• Deploy plugin to production
• Run initial batch embedding (off-peak hours)
• Verify all connections working
• Test sample recommendations on live site
• Enable monitoring dashboards
• Announce to content team

☑️ Post-Launch (First Week)

• Monitor error logs daily
• Track API usage and costs
• Review recommendation acceptance rates
• Gather user feedback from content team
• Adjust similarity thresholds if needed
• Document any issues and resolutions

☑️ Ongoing (Monthly)

• Review analytics dashboard
• Optimize business rules based on data
• Monitor vector DB storage usage
• Review and optimize API costs
• Update documentation as needed
• Plan feature enhancements