██FR█████ █INTELL███████████

Learn with Daily Mastery – From AI Generation to Email Delivery


This content originally appeared on DEV Community and was authored by Dale Nguyen

One of the biggest challenges in online education isn’t creating content—it’s delivering the right content to the right person at the right time. When building DailyMastery, we needed to solve a complex scheduling problem: how do you deliver personalized, AI-generated lessons to users across different time zones, preferences, and learning schedules?

Our solution is a high-performance lesson scheduling system that processes thousands of users daily, generates personalized content on-the-fly, and delivers beautiful email lessons at precisely the right moment. Here’s how we built it.

DailyMastery email

The Challenge: Personalization at Scale

Traditional learning platforms send the same content to everyone. But true personalization requires:

  • Individual content generation: Each lesson tailored to specific goals and progress
  • Time-based delivery: Respecting user preferences for morning, afternoon, or evening learning
  • Scalable processing: Handling growing user bases without performance degradation
  • Reliability: Ensuring lessons are delivered consistently, even when individual components fail

Architecture Overview: From User Preferences to Inbox

Our lesson scheduling system consists of several coordinated services:

1. User Selection Engine

The system starts by identifying which users should receive lessons for each time slot:

Selection Criteria:

  • Users must have preferences.studyTime matching the time slot (‘morning’, ‘afternoon’, ‘evening’)
  • Users must have preferences.notifications.dailyReminders enabled
  • Users must have active study plans with lessons scheduled for today

Time Slots:

  • 06:00 UTC (Morning): Targets users who prefer early learning
  • 12:00 UTC (Afternoon): Catches lunch-break learners
  • 18:00 UTC (Evening): Reaches after-work studiers

2. AI Content Generation Pipeline

Once we identify users, each lesson goes through our AI enhancement process:

Process Flow:

  1. Lesson Retrieval: Get scheduled lesson from database with original template content
  2. AI Enhancement: Use Google Cloud Vertex AI with Gemini to expand and personalize the content
  3. Quality Validation: Ensure AI-generated content meets quality standards
  4. Fallback Strategy: Use original template if AI generation fails

Content Personalization:
The AI considers user context including:

  • Learning objectives and goals
  • Current progress and difficulty level
  • Preferred learning style and time commitment
  • Previous lesson feedback and engagement

3. Batch Processing for Scale

To handle thousands of users efficiently, we use sophisticated batch processing:

Concurrency Control:

const pLimit = (await import('p-limit')).default
const limit = pLimit(10) // Process up to 10 users concurrently

const promises = userGroups.map((userGroup) => limit(() => this.processUserGroup(userGroup)))

await Promise.allSettled(promises)

Error Resilience:
Individual user failures don’t stop the entire batch. Each user’s processing is isolated, logged, and reported separately.

The Lesson Processor: Heart of the System

Our LessonProcessorService handles the complex task of managing lesson content:

Active Plan Discovery

The system performs comprehensive scans to find all active study plans:

async getActivePlans(): Promise<UserPlanGroup[]> {
  // Scan all users with active plans
  // Find today's scheduled lessons
  // Group by user for efficient processing
}

AI Content Generation

Each lesson is enhanced with AI-generated content:

Quality Metrics:

  • Content length and depth analysis
  • Learning objective alignment
  • Engagement potential scoring
  • Readability assessment

Performance Tracking:

  • AI generation time per lesson
  • Success/failure rates
  • Content quality improvement over time
  • User engagement correlation

Database Integration

Seamless integration with Firestore for lesson management:

  • Atomic updates for lesson content
  • Timestamp tracking for delivery status
  • Progress tracking for user analytics
  • Batch operations for efficiency

Email Delivery: The Final Mile

Our EmailCoordinatorService handles the critical task of email delivery:

Personalized Templates

Each email is personalized based on:

  • Time-based greetings: 🌅 “Good morning” vs ☀ “Good afternoon” vs 🌙 “Good evening”
  • Study plan context: Extracted plan names and goals
  • Progress indicators: Current lesson number and completion status
  • Engagement elements: Interactive exercises and progress tracking

Responsive Design

Our TemplateEngineService creates mobile-friendly emails with:

  • Maximum width of 600px for optimal display
  • Modern gradient designs and professional styling with CSS animations
  • Clear call-to-action buttons for lesson engagement
  • Progress visualization and success metrics
  • Time-based personalized greetings and content adaptation
  • Professional email templates with comprehensive fallback support

Delivery Reliability

Email delivery includes comprehensive error handling:

  • Detailed error logging for failed deliveries
  • Retry mechanisms for transient failures
  • Admin alerting for critical delivery issues
  • Success rate monitoring and reporting
  • Batch email API with intelligent fallback to individual sends
  • Rate limiting protection (2 requests per second compliance)

Real-Time Processing Flow

Here’s what happens when our scheduler processes the morning time slot:

1. Trigger Processing (06:00 UTC) 

POST /schedule/06:00

2. User Discovery

  • Query database for morning-preference users
  • Filter for active daily reminders
  • Identify today’s scheduled lessons

3. Batch Processing

  • Group users for efficient processing
  • Apply concurrency limits (10 concurrent users)
  • Process each user’s lessons independently

4. Content Generation

For each user:

  • Retrieve scheduled lessons from database
  • Generate AI-enhanced content
  • Update lesson records with new content
  • Mark lessons as processed

5. Email Delivery

  • Render personalized HTML templates
  • Send via Resend email service
  • Track delivery status and metrics
  • Update lesson records with sent timestamps

6. Reporting and Monitoring

  • Generate processing summary
  • Send admin reports via email
  • Log performance metrics
  • Alert on critical errors

Performance Metrics and Monitoring

Our system tracks comprehensive metrics:

Processing Performance

  • Total processing time per time slot
  • Average time per user
  • Concurrency efficiency metrics
  • Database operation latency

Content Quality

  • AI generation success rates (target: >95%)
  • Content length and complexity metrics
  • User engagement correlation with AI content
  • Fallback usage rates

Email Delivery

  • Email send success rates (target: >99%)
  • Template rendering performance
  • User engagement rates (opens, clicks, completions)
  • Bounce and complaint rates

Future Enhancements

We’re continuously improving the scheduling system:

Advanced Personalization

  • Learning pattern analysis: Adapt timing based on user engagement patterns
  • Content difficulty adjustment: Dynamic complexity scaling based on progress
  • Multi-modal delivery: Support for SMS, push notifications, and in-app delivery

Performance Optimization

  • Predictive scaling: Auto-scaling based on user growth patterns
  • Content pre-generation: Cache frequently used AI-generated content
  • Global distribution: Multi-region deployment for reduced latency

Analytics and Insights

  • Learning effectiveness metrics: Correlation between delivery timing and retention
  • Content performance analysis: A/B testing for AI-generated vs template content
  • User journey optimization: Personalized scheduling based on individual success patterns

Lessons Learned

Building a personalized lesson scheduling system taught us:

Technical Insights

  1. Batch processing with concurrency limits provides the best balance of speed and reliability
  2. Comprehensive error handling at every level is essential for user trust
  3. AI content generation requires robust fallback strategies for reliability
  4. Email delivery is more complex than it appears – template rendering, personalization, and deliverability all matter

Product Insights

  1. Time-based personalization significantly improves engagement compared to generic scheduling
  2. AI-enhanced content shows measurably higher completion rates than template-based lessons
  3. Processing transparency builds user confidence – users appreciate knowing when their lessons are being prepared
  4. Reliable delivery is table stakes – users expect lessons to arrive consistently

Getting Started

If you’re building your own personalized delivery system, consider:

  1. Start with reliable basics: Get consistent delivery working before adding AI enhancement
  2. Design for failure: Assume every external service will fail and plan accordingly
  3. Monitor everything: Comprehensive logging and metrics are essential for debugging and optimization
  4. Test with real load: Batch processing behaves differently at scale

Our lesson scheduling system demonstrates that personalized, AI-powered education delivery is not only possible but can be built reliably and efficiently with the right architecture and attention to detail.


This content originally appeared on DEV Community and was authored by Dale Nguyen