CONFIDENCE LEVEL: ___/10

If score < 8/10 confidence → Iterate with refined prompt

• Quick Validation Commands:

Check slide count (after download, macOS):

mdls -name kMDItemNumberOfPages "chapter-XX-slides.pdf"

Verify themes coverage (manual):

• Open PDF and scan slide titles
• Check each of 5-7 numbered themes from prompt appears
• Flag any missing themes for iteration

Confidence Scoring Guide:

• 9-10/10: All 7 criteria pass, themes fully covered, tone perfect → Deploy

• 7-8/10: 5-6 criteria pass, minor tone/format adjustments needed → Consider iteration

• 5-6/10: 3-4 criteria pass, significant iteration required → Refine prompt

• <5/10: Start over with refined prompt focusing on failed dimensions

• Iterate if needed: If slides miss the mark, click "Edit" and refine prompt with more explicit constraints focusing on failed criteria

5. Deploy (3 min)

• Click "Download" button (appears after generation)
• PDF downloads to your browser's default location

File Naming Standard:

• Format: chapter-{NN}-slides.pdf where NN is zero-padded
• Examples:
  • Chapter 1 → chapter-01-slides.pdf (NOT chapter-1-slides.pdf)
  • Chapter 12 → chapter-12-slides.pdf
  • Chapter 31 → chapter-31-slides.pdf
• Why zero-padding: Ensures correct alphabetical sorting in file systems

Renaming & Moving:

# Example for Chapter 1:
mv ~/Downloads/"The-AI-Coding-Revolution.pdf" "apps/learn-app/static/slides/chapter-01-slides.pdf"

# Example for Chapter 31:
mv ~/Downloads/"Designing-Reusable-Intelligence.pdf" "apps/learn-app/static/slides/chapter-31-slides.pdf"

• Integration Path: All slides must be in apps/learn-app/static/slides/ directory

6. Integrate into Documentation (2 min)

• Open chapter README: apps/learn-app/docs/[part]/[chapter]/README.md
• Add slides metadata to frontmatter (YAML at top of file):

---
sidebar_position: 1
title: "Chapter 1: [Chapter Title]"
slides:
  source: "slides/chapter-01-slides.pdf"
  title: "Chapter 1: [Chapter Title]"
  height: 700
---

# Chapter 1: [Chapter Title]

[Chapter introduction content...]

## 🎯 Before You Begin

---

## What You'll Learn

By the end of this chapter, you'll understand:
- [Learning objectives...]

Integration Details:

• Location: Add slides: metadata in frontmatter YAML block at top of file
• Format: Use object format with source, title, and height fields
• Source Path: Use relative path slides/chapter-XX-slides.pdf (NOT absolute path /slides/...)
• Automatic Injection: Build-time plugin automatically injects PDFViewer component BEFORE "What You'll Learn" heading
• No Manual Import: Do NOT add import PDFViewer or JSX components (handled automatically)
• Placement: Slides appear after "## 🎯 Before You Begin" and before "## What You'll Learn"
• Rationale: Progressive disclosure - students see big picture (slides) before detailed objectives

Alternative Simple Format (for quick integration):

---
title: "Chapter Title"
slides: "slides/chapter-01-slides.pdf"
---

This uses default height (700) and default title ("Chapter Slides").

Verification:

• Save file and build Docusaurus: npm run build
• Check build output for: [Slides Transformer] ✅ Injected slides/chapter-XX-slides.pdf
• Verify slides appear correctly in deployed site

Metadata-Driven Architecture (NEW)

What Changed: The integration workflow now uses metadata-driven slide injection instead of manual JSX components.

Old Approach (Deprecated):

import PDFViewer from '@site/src/components/PDFViewer';

<PDFViewer src="slides/chapter-01-slides.pdf" title="..." height={700} />

New Approach (Current):

---
slides:
  source: "slides/chapter-01-slides.pdf"
  title: "Chapter 1: Title"
  height: 700
---

Benefits:

• ✅ Cleaner Content: No JSX imports or components cluttering markdown
• ✅ Consistent Placement: Automatic injection before "What You'll Learn" (no manual positioning)
• ✅ Cloud-Ready: Supports both local paths (slides/file.pdf) and URLs (https://cdn.example.com/file.pdf)
• ✅ Easier Maintenance: Update metadata, not JSX code
• ✅ Build-Time Processing: Remark plugin handles transformation automatically

How It Works:

1. Frontmatter metadata specifies slides source, title, and height
2. Build-time remark plugin reads frontmatter during Docusaurus compilation
3. Plugin finds "What You'll Learn" heading in markdown AST
4. Plugin injects PDFViewer component node before that heading
5. Docusaurus compiles MDX with component already in place
6. Result: Clean HTML with slides embedded at correct location

For Cloud Migration: When PDFs move to Cloudflare R2/S3, simply update the source field to the URL:

slides:
  source: "https://r2.cloudflare.com/tutorsgpt/slides/chapter-01.pdf"
  title: "Chapter 1: Title"
  height: 700

No code changes needed—just metadata updates.

Time Estimates by Experience Level

Efficiency gains: After 2-3 chapters, you'll develop muscle memory for prompt engineering and can reuse/adapt previous successful prompts.

Time Variability Notes:

• Generation time varies widely (5-35 minutes per chapter)
• Complex chapters with more sources take longer
• Daily limits may require splitting work across multiple days
• Batch processing saves ~25% time through parallel preparation/generation

🚨 Daily Limit Troubleshooting

Symptom: "You have reached your daily Slides limits" message appears after clicking "Generate"

What this means:

• NotebookLM enforces per-account daily slide generation limits
• Limit resets at midnight Pacific Time (Google servers)
• Notebooks remain available; only slide generation is blocked
• Exact limit unknown (observed: 3-5 chapters per day depending on complexity)

Recovery Options:

1. Wait Method: Return after 24 hours and continue from Step 4 (notebooks persist)
2. Account Switching: Use different Google account if available (requires re-creating notebooks)
3. Batch Planning: Plan to generate 3-5 chapters per day maximum across multiple days

Prevention Strategy:

• Track chapters completed per day in project progress tracker
• Schedule batch work across multiple days for large parts (e.g., Part 4 with 18 chapters = 4-6 days)
• Generate slides for highest-priority chapters first each day

If blocked mid-batch:

• Notebooks with uploaded sources are saved
• Resume slide generation next day starting at Step 2 (Configure Slide Deck)
• No need to re-upload sources

Batch Processing Multiple Chapters

For processing 3+ chapters in one session (recommended for Parts with multiple chapters):

Preparation Phase (10-15 min)

1. Create all notebooks with titles FIRST (before any slide generation)
2. Upload sources for all chapters (notebooks persist independently)
3. Prepare all prompts in text editor (adapt from proficiency-calibrated templates)
4. Review chapter-index.md to confirm proficiency levels match templates

Generation Phase (per chapter, 15-30 min each)

5. Generate slides for Chapter N
6. While Chapter N generates, prepare/refine prompt for Chapter N+1 in text editor
7. Download Chapter N PDF when ready
8. Start Chapter N+1 generation immediately
9. Rename/move Chapter N PDF while N+1 generates (parallel work)

Integration Phase (bulk, 10-15 min)

10. Move all PDFs to apps/learn-app/static/slides/ at once
11. Add slides metadata to frontmatter in all chapter READMEs sequentially
12. Verify all slides load correctly in local Docusaurus build (npm run build)

Time Savings

• Sequential approach: 3 chapters × 25 min = 75 min
• Batch approach: Setup (15 min) + Generation (45 min overlapped) + Integration (15 min) = ~60 min
• Savings: ~25% reduction through parallel preparation/generation

Daily Limit Awareness

• Plan for 3-5 chapters maximum per day
• If limit hit: Stop, update progress tracker, resume next day
• For large parts (e.g., Part 4 with 18 chapters): Schedule 4-6 days of work

Automation via Playwright MCP (Optional Advanced)

For batch processing multiple chapters with browser automation:

Prerequisites

• Playwright MCP server installed and configured in Claude Code
• Google account credentials ready
• Access to NotebookLM sources (markdown files or deployed URLs)

Workflow Pattern (per chapter)

1. Navigate to notebooklm.google.com and authenticate
2. Create notebook with standardized title
3. Upload sources via file upload dialog (automate file selection)
4. Navigate to "Slide Deck" in Studio panel
5. Configure format (Presenter Slides, Default/Long length)
6. Paste prompt into customization field
7. Click "Generate" and monitor (5-30 min wait)
8. Detect completion (download button appears)
9. Download PDF to .playwright-mcp/ directory
10. Move PDF to apps/learn-app/static/slides/ with standardized naming
11. Update chapter README frontmatter with slides metadata (automated text manipulation)

Benefits

• Time Savings: ~40% faster for 3+ chapters due to reduced context switching
• Consistency: Standardized naming, no manual typos
• Automation: Can queue multiple chapters

Limitations

• Technical Setup Required: MCP server configuration, browser automation knowledge
• Daily Limits Still Apply: Cannot bypass NotebookLM's per-account limits
• Extended Wait Times: Require patient monitoring (30+ min generations)
• Error Handling: Must handle stuck generations, timeouts, UI changes

When to Use

• Processing 5+ chapters in batch
• Repetitive workflow across similar chapters
• Team environment with automation infrastructure

When NOT to Use

• First 1-2 chapters (learn manual workflow first)
• Chapters requiring significant prompt iteration
• One-off slide generation

Required Behaviors (Positive Framing)

Follow these essential practices for high-quality slide generation:

✅ Audience Specification

• DO: Use specific proficiency-calibrated audience descriptions with prerequisites
  • Example: "A2 beginners with no programming experience discovering AI-native development"
• WHY: Enables NotebookLM to calibrate tone, complexity, and vocabulary appropriately

✅ Framework Alignment

• DO: Explicitly state your educational framework with 3-5 core principles
  • Example: "AI-native development: AI is your core partner (not just a helper)"
• WHY: NotebookLM doesn't know your teaching philosophy unless you make it explicit

✅ Concrete Examples

• DO: Include specific data, facts, and numbers in your numbered themes
  • Example: "$3 trillion developer economy transforming in 2-3 years"
• WHY: Specificity makes abstract concepts tangible and memorable

✅ Multi-Faceted Tone

• DO: Describe tone with contrasts and context
  • Example: "Encouraging (not intimidating) — beginners may feel overwhelmed"
• WHY: Single-word descriptors are ambiguous; context enables accurate calibration

✅ Explicit Format Constraints

• DO: Specify slide count range, bullets per slide, and use XML tags
  • Example: <slide_format_requirements>Generate exactly 12-15 slides...</slide_format_requirements>
• WHY: Prevents text-heavy lecture notes; ensures presenter-friendly slides

✅ Narrative Progression

• DO: Define clear narrative arc with problem → transformation → opportunity → action
  • Example: "Start with current pain points, show transformation, end with next steps"
• WHY: Pedagogical progression builds understanding incrementally

✅ Actionable Endings

• DO: End with concrete, specific next steps
  • Example: "Complete Chapter 2 exercises, set up development environment, join Discord"
• WHY: Vague summaries ("Keep learning!") provide no guidance for student action

Creative Variance

NotebookLM tends to:

• Default to academic/lecture tone even when asked for accessible style
• Create text-heavy slides despite "visual" requests
• Use generic titles instead of engaging ones
• Skip the emotional/motivational framing

Counter this by:

• Reinforcing tone with contrasts: "Encouraging (not academic)"
• Explicitly requesting "3-5 bullet points per slide (not paragraphs)"
• Providing example of engaging title style in prompt
• Including "opportunity-driven" or "transformation-focused" in tone

Vary slide structure: Don't always use bullet points. Request timelines, comparison tables, process flows, or numbered steps when appropriate to the content.

Concrete Example: Chapter 1 AI Development (A2)

Prompt Used:

Create an inspiring slide deck for absolute beginners (A2 proficiency level) learning AI-driven software development.

AUDIENCE: Complete beginners with no prior programming experience discovering that AI has fundamentally changed how software is built.

AI-NATIVE DEVELOPMENT TO EMPHASIZE:
• AI as Core Partner: AI is your primary development partner, not just a helper tool
• Role Transformation: From "typist" (writing syntax) to "orchestrator" (directing AI)
• Spec-Driven Development: You write specifications; AI generates implementations
• Co-Learning: Humans and AI learn together through iteration

KEY THEMES TO HIGHLIGHT:
1. The $3 trillion developer economy transformation happening NOW
2. Why this moment is an OPPORTUNITY, not a threat (AI lowers barriers)
3. The shift from syntax memorization to specification writing and system design
4. Evolution through 4 generations: code completion → chat → agentic coding → autonomous agents
5. What makes you valuable: judgment, architecture, domain expertise, orchestration skills

TONE & STYLE:
• Encouraging and empowering (not intimidating) — these slides will be presented to absolute beginners who may feel overwhelmed by AI development, so maintaining an encouraging tone is critical to prevent dropout
• Future-focused and opportunity-driven — frame AI as opening doors, not closing them
• Clear, simple language avoiding jargon — learners have no technical background yet
• Action-oriented: what students should DO next

<slide_format_requirements>
Generate exactly 12-15 slides total. Each slide must contain 3-5 concise bullet points written as complete sentences, NOT dense paragraph text. Use visual hierarchy with clear headings. This slide deck should comprehensively cover the chapter content using your full output context to create a thorough, well-structured presentation. Don't artificially limit scope—make sure you address all major themes before concluding.
</slide_format_requirements>

NARRATIVE ARC:
• Progressive structure: problem → transformation → opportunity → action
• End with clear, actionable next steps for learners (specific tasks, not vague encouragement)

Result: "The AI Coding Revolution" - 13 slides, all themes covered, encouraging tone, actionable ✅

Success Indicators

Working:

• ✅ Title matches framework ("AI Coding Revolution" reflects AI-native philosophy)
• ✅ Proficiency-appropriate (simple vocabulary for A2, no jargon)
• ✅ All 5 themes covered with specific data ($3T economy, 4 generations)
• ✅ Tone consistent (encouraging, not academic)
• ✅ Format followed (13 slides, 3-5 bullets each)
• ✅ Clear arc (problem → transformation → opportunity → action)
• ✅ Actionable ending (concrete next steps)

Failed:

• ❌ Generic title ("Introduction to AI Development")
• ❌ Text-heavy paragraphs instead of bullets
• ❌ Missing numbered themes
• ❌ Academic tone for beginners
• ❌ Random sequence without progression
• ❌ Vague takeaways ("Keep learning!")