AI Course Creation Pipeline - Architecture & Design
Vision
Transform a vague course idea into a complete, world-class educational product through AI-powered research, planning, and content generation.
User Journey
User: "I want to create a course on AI agents for beginners"
↓
System: Refines → Researches → Plans → Generates → Publishes
↓
Output: Complete course with lessons, videos, blogs, exercises
System Architecture
Phase 1: Ideation & Prompt Refinement
Goal: Transform vague ideas into structured research prompts
Input: Rough course idea
- "AI agents course"
- "Teach people about LangChain"
- "Advanced prompt engineering"
AI Processor:
- Clarification questions (what, who, why, level, format)
- Prompt engineering to create world-class research query
- Define success criteria and scope
Output: Structured research brief
{
"course_idea": "Building Autonomous AI Agents with Claude",
"target_audience": "Intermediate developers with Python experience",
"learning_outcomes": ["Build agents that use tools", "Implement memory systems"],
"research_questions": [
"What are current best practices for agent architectures?",
"What tools/frameworks are most popular in 2025?",
"What gaps exist in current educational content?"
],
"competitive_landscape": "Analyze existing courses from DeepLearning.AI, OpenAI, Anthropic",
"unique_angle": "Focus on production-ready patterns, not toy examples"
}
Phase 2: Research & Context Gathering
Goal: Build comprehensive knowledge base about the topic
Research Sources:
-
Tavily API (Real-time web search)
- Latest trends and news
- Academic papers
- Industry blog posts
- Best practices
-
Google Search (Broad coverage)
- Existing courses and tutorials
- Market analysis
- Popular frameworks and tools
- Community discussions
-
Firecrawl (Deep content extraction)
- Extract full content from top resources
- Documentation scraping
- Tutorial extraction
- Code example mining
-
LLM Synthesis (Knowledge integration)
- Summarize findings
- Identify patterns
- Gap analysis
- Competitive positioning
Output: Research Report
{
"market_analysis": {
"existing_courses": [...],
"gaps": [...],
"trends": [...]
},
"technical_landscape": {
"popular_frameworks": ["LangChain", "AutoGPT", "Claude SDK"],
"key_concepts": ["Agents", "Tools", "Memory", "Planning"],
"code_examples": [...]
},
"pedagogical_insights": {
"what_works": [...],
"common_mistakes": [...],
"learner_pain_points": [...]
},
"content_opportunities": {
"unique_angles": [...],
"differentiators": [...],
"value_propositions": [...]
}
}
Phase 3: Course Planning & Architecture
Goal: Design comprehensive course structure using research context
Inputs:
- Research report from Phase 2
- User preferences (duration, depth, format)
- Pedagogical best practices
AI Course Architect:
- Apply Bloom's Taxonomy
- Design learning progression
- Plan content types (lessons, videos, hands-on projects)
- Define assessment strategy
Output: Course Blueprint
{
"course_metadata": {
"title": "Building Production-Ready AI Agents with Claude",
"description": "...",
"difficulty": "intermediate",
"total_duration_hours": 12
},
"sections": [
{
"title": "Agent Fundamentals",
"lessons": [
{
"title": "What Are Autonomous Agents?",
"type": "theory",
"duration_minutes": 30,
"deliverables": ["lesson_content", "video_script", "blog_post"]
}
]
}
],
"content_calendar": {
"lessons_to_write": 20,
"videos_to_create": 15,
"blog_posts_to_write": 8,
"exercises_to_design": 25
}
}
Phase 4: Content Generation
Goal: Create all course materials
Content Types:
-
Lesson Content (Markdown)
- Use existing world-class prompts
- 2000-4000+ words per lesson
- Code examples and exercises
- Special blocks (warnings, tips, quizzes)
-
Video Scripts
- Scene-by-scene breakdown
- Visual cues and timestamps
- Code demonstrations
- Talking points
-
Blog Posts
- SEO-optimized content
- Companion pieces to lessons
- Tutorial format
- Social media snippets
-
Code Examples
- Progressive complexity
- Full working examples
- GitHub repository structure
- Unit tests
-
Exercises & Assessments
- Hands-on challenges
- Quizzes
- Projects
- Success criteria
Output: Complete Content Package
Phase 5: Media & Asset Generation
Goal: Create supporting materials
-
Video Production
- Scripts with timestamps
- Code demonstration flows
- Screen recording guides
- Animation suggestions
-
Visual Assets
- Diagrams (architecture, flow charts)
- Infographics
- Presentation slides
- Thumbnail designs
-
Interactive Elements
- Code playgrounds
- Live demos
- Jupyter notebooks
- API sandboxes
Technical Implementation
Technology Stack
Backend Services:
- Tavily API: Real-time web search ($50/month for 1000 searches)
- Firecrawl: Content extraction ($20/month starter)
- OpenRouter: Multi-LLM access (GPT-4, Claude, Gemini)
- Supabase: Database and storage
Agent Framework:
- LangGraph or Claude Agent SDK for multi-step workflows
- Tavily Tool: Web search integration
- Firecrawl Tool: Content extraction
- Custom Tools: Research synthesis, outline generation
Frontend Components:
- Course Planner UI: New route
/admin/course-planner - Research Dashboard: View research findings
- Content Generator: Batch content creation
- Progress Tracker: Track course creation pipeline
User Interface Flow
1. Course Planner Page (/admin/course-planner)
┌─────────────────────────────────────────────────┐
│ 🎓 AI Course Creation Pipeline │
├─────────────────────────────────────────────────┤
│ │
│ Step 1: Describe Your Course Idea │
│ ┌────────────────────────────────────────────┐ │
│ │ I want to create a course on... │ │
│ │ │ │
│ │ │ │
│ └────────────────────────────────────────────┘ │
│ │
│ [Refine My Idea] ────────────────────────────→ │
│ │
├─────────────────────────────────────────────────┤
│ Step 2: Research & Context (AI Agent Working) │
│ ┌────────────────────────────────────────────┐ │
│ │ ⏳ Searching web for latest trends... │ │
│ │ ✅ Found 24 relevant articles │ │
│ │ ⏳ Extracting content from top resources │ │
│ │ ⏳ Analyzing competitive landscape │ │
│ └────────────────────────────────────────────┘ │
│ │
│ [View Research Report] │
│ │
├─────────────────────────────────────────────────┤
│ Step 3: Review Course Plan │
│ ┌────────────────────────────────────────────┐ │
│ │ Course: Building AI Agents with Claude │ │
│ │ • 5 sections, 20 lessons │ │
│ │ • 12 hours total │ │
│ │ • Intermediate level │ │
│ │ │ │
│ │ [Edit Outline] [Approve & Generate] │ │
│ └────────────────────────────────────────────┘ │
│ │
├─────────────────────────────────────────────────┤
│ Step 4: Content Generation │
│ ┌────────────────────────────────────────────┐ │
│ │ ⏳ Generating lesson 1/20... │ │
│ │ ✅ Lesson content created (3,245 words) │ │
│ │ ✅ Video script generated │ │
│ │ ✅ Blog post drafted │ │
│ │ │ │
│ │ Progress: ████████░░░░░░░░ 40% │ │
│ └────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────┘
Implementation Phases
✅ Phase 1.1: Ideation System (Week 1)
- Build prompt refinement engine
- Create clarification question system
- Generate structured research brief
- UI: Course idea input form
🔄 Phase 1.2: Research Orchestration (Week 2)
- Set up Tavily API integration
- Set up Firecrawl API integration
- Build research agent with LangGraph
- Create research report template
- UI: Research progress dashboard
⏳ Phase 2: Course Planning (Week 3)
- Integrate world-class course outline prompt
- Build course architect agent
- Create content calendar generator
- UI: Course outline editor
⏳ Phase 3: Content Generation (Week 4)
- Batch lesson generation
- Video script generator
- Blog post generator
- Code example generator
- UI: Content generation dashboard
⏳ Phase 4: Asset Generation (Week 5)
- Diagram generator
- Slide deck creator
- Interactive element designer
API Requirements
Tavily API
# Installation
npm install @tavily/core
# Usage
const tavily = new TavilyClient({ apiKey: process.env.TAVILY_API_KEY });
const results = await tavily.search("AI agent best practices 2025");
Pricing: $50/month for 1000 searches
Firecrawl
# Installation
npm install @mendable/firecrawl-js
# Usage
const firecrawl = new FirecrawlApp({ apiKey: process.env.FIRECRAWL_API_KEY });
const content = await firecrawl.scrapeUrl("https://docs.anthropic.com");
Pricing: $20/month starter plan
Cost Estimation
Monthly Costs:
- Tavily API: $50
- Firecrawl: $20
- OpenRouter (LLM calls): $30-50
- Total: ~$100-120/month
Per Course Creation:
- Research phase: ~$5 (10-20 searches + extractions)
- Planning phase: ~$2 (LLM calls)
- Content generation (20 lessons): ~$20 (LLM calls)
- Total per course: ~$25-30
Success Metrics
-
Time Savings
- Traditional: 40-60 hours per course
- AI-Powered: 8-12 hours per course
- Savings: 75-80%
-
Quality Metrics
- Pedagogical alignment: Bloom's Taxonomy applied
- Content depth: 2000-4000+ words per lesson
- Code examples: 4-6+ per lesson
- Research-backed: Data from multiple sources
-
Output Completeness
- ✅ Course outline
- ✅ All lesson content
- ✅ Video scripts
- ✅ Blog posts
- ✅ Code examples
- ✅ Exercises
Future Enhancements
- Voice Generation: Convert scripts to audio
- Video Generation: Auto-create videos from scripts
- Translation: Multi-language course generation
- Personalization: Adapt content to learner level
- Auto-Update: Keep courses current with latest trends
Next Steps
- ✅ Review this architecture - Confirm approach
- 🔄 Build Phase 1.1 - Ideation system
- ⏳ Set up APIs - Tavily, Firecrawl accounts
- ⏳ Create UI - Course Planner page
- ⏳ Implement agents - Research orchestrator