The sophisticated content-aware AI system has been fully restored after fixing syntax errors.

• Multi-Slide Content-Aware AI: AI sees ALL slides and their text elements with context
• Multi-Slide Operations: AI can edit multiple slides in one request ("make whole deck about cats")
• Text Box Relationship Understanding: AI knows which elements belong to the same text box
• Smart Slide Targeting: AI automatically finds the right slide based on content
• Text Box Replacement: AI can replace entire text boxes (all bullets) or individual elements
• Markdown to PowerPoint XML: AI generates clean markdown, system converts to proper XML
• Precise Element Selection: AI chooses the right text element using slide-aware IDs (1A, 2B, 3C, etc.)
• Natural Language Commands: "Change the executive summary to something about cats"
• Surgical XML Editing: Makes precise edits to specific elements without corruption
• File Upload/Download: Drag & drop PPTX processing
• Thread-Based Chat: Multiple conversation threads with persistent history

• Syntax Errors: Resolved all missing braces and orphaned code blocks
• Content-Aware System: Fully restored the sophisticated AI processing
• Element Mapping: Restored precise element targeting and surgical edits

"Change the main heading on the first slide to 5 words about tortoises"
"Replace the title with something about giraffes"
"Update the large text to say hello world"
"Add some lorem ipsum to the slide"

1. Content Extraction: System reads slide XML and extracts all text elements with context
2. AI Analysis: OpenAI sees actual slide content: [A] "Kwaku Slide One" (18pt, top-center)
3. Smart Selection: AI chooses the right element based on user's natural language
4. Precise Edit: System makes surgical XML changes to exactly the right element

interface SlideText {
  text: string;           // "Welcome to Company"
  fontSize: number;       // 44
  position: {...};        // top-center
  elementId: string;      // "A"
  xmlPath: string;        // For precise editing
}

PPTX Content:

Slide 1 text elements:
Text Box TB0 (title):
  [1A] "Welcome to Company" (44pt, top-center)
Text Box TB1 (body):
  [1B] "Quarterly Review" (18pt, middle-left)

Slide 2 text elements:
Text Box TB0 (title):
  [2A] "Executive Summary" (36pt, top-center)
Text Box TB1 (body):
  [2B] "Key findings and recommendations" (16pt, middle-left)
  [2C] "First bullet point" (16pt, middle-left)
  [2D] "Second bullet point" (16pt, middle-left)

User: "Change the executive summary content to 3 bullet points about cats"
AI Decision: "User wants to replace entire body content under Executive Summary"
AI Response: {
  "textBoxId": "TB1", 
  "slideNumber": 2,
  "content": "• Cats are independent animals\n• Cats love to sleep in sunny spots\n• Cats communicate through purring"
}
System: Parses markdown → Generates proper PowerPoint XML → Replaces entire text box

The system uses deep knowledge of PPTX structure to make precise edits. See README-how_pptx-works.md for complete technical details on:

• PPTX ZIP structure and relationships
• XML parsing and text extraction
• Safe editing patterns that preserve file integrity

Understanding how PowerPoint organizes content is crucial for the AI system to make intelligent editing decisions.

• Text Box = <p:sp> element (PowerPoint shape container)
• Text Elements = <a:t> tags within the text box (individual text runs)
• One text box can contain multiple paragraphs (<a:p>) with multiple text runs (<a:t>)

<!-- Executive Summary slide has TWO text boxes: -->

<!-- Text Box 1: Title -->
<p:sp><p:ph type="title"/>
  <a:t>Executive Summary</a:t>
</p:sp>

<!-- Text Box 2: ALL bullets in ONE text box -->
<p:sp><p:ph type="body"/>
  <a:p><a:buChar char="•"/><a:t>First bullet point</a:t></a:p>
  <a:p><a:buChar char="•"/><a:t>Second bullet point</a:t></a:p>
  <a:p><a:buChar char="•"/><a:t>Third bullet point</a:t></a:p>
</p:sp>

• Preserves XML Structure: Each <a:t> replacement maintains bullet formatting
• Avoids Corruption: No need to reconstruct complex XML with indentation, bullet chars, etc.
• Granular Control: Can replace individual bullets or groups of bullets
• User Flexibility: "Change first bullet" vs "change all executive summary content"

• <p:ph type="title"/> - Main slide titles
• <p:ph type="body"/> - Content areas (usually containing bullets)
• <p:ph type="ctrTitle"/> - Centered titles
• <p:ph type="subTitle"/> - Subtitles

The AI sees individual elements but understands their relationships:

Text Box 1 (title): [2A] "Executive Summary"
Text Box 2 (body): [2B] "First bullet" [2C] "Second bullet" [2D] "Third bullet"

When user says "change executive summary content" → AI targets the entire text box When user says "change executive summary title" → AI targets element 2A

The system uses a clean separation between content generation and technical implementation:

1. AI generates markdown-like content:

   • First bullet point about cats
   • Second bullet point about dogs  
   • Third bullet point about birds
   

2. System parses and converts to PowerPoint XML:

   <p:txBody>
     <a:p><a:pPr><a:buChar char="•"/></a:pPr><a:t>First bullet point about cats</a:t></a:p>
     <a:p><a:pPr><a:buChar char="•"/></a:pPr><a:t>Second bullet point about dogs</a:t></a:p>
     <a:p><a:pPr><a:buChar char="•"/></a:pPr><a:t>Third bullet point about birds</a:t></a:p>
   </p:txBody>

Benefits:

• ✅ AI focuses on content: No need to understand PowerPoint XML complexity
• ✅ System handles formatting: Proper bullet structure, indentation, XML escaping
• ✅ Clean separation: Content generation vs. technical implementation
• ✅ Flexible: Supports bullets, sub-bullets, plain text, mixed content

• AI sees actual slide content (not blind heuristics)
• Smart element selection based on context
• Natural language commands work reliably
• Precise XML editing without file corruption
• Clean, maintainable codebase under 500 lines per file

├── backend/
│   ├── index.ts                 # Routes & database (~200 lines)
│   ├── pptx.ts                  # Content-aware PPTX processing (~400 lines)  
│   └── ai.ts                    # AI command parsing (~150 lines)
├── frontend/
│   └── index.html              # Simple web interface
└── README.md                   # This file

• Content-Aware: AI sees actual slide content before making decisions
• Precise Editing: Surgical XML changes to specific elements
• Clean Code: All files under 500 lines, clear separation of concerns
• No Legacy Cruft: Removed all broken fallback code
• Maintainable: Easy to find, understand, and modify functionality

• GET / - Main chat interface
• GET /health - Health check
• GET /api/chat - Get chat history
• POST /api/chat - Send message (with content-aware AI)
• POST /api/upload - Upload PPTX file
• GET /api/download/:fileId - Download processed file

• File Size Limit: Keep files under 500 lines for maintainability
• Single Responsibility: Each file has one clear purpose
• No Legacy Cruft: Remove broken code instead of keeping "fallbacks"
• Content-Aware First: AI should see actual content, not use blind heuristics

• Keep README Updated: When you make code changes, update this README to reflect the current state
• Technical Reference: See README-how_pptx-works.md for detailed PPTX XML structure documentation
• Document Milestones: Update status sections when major features are completed
• Remove Outdated Content: Delete old implementation details that no longer apply

1. Intelligence Over Heuristics: Let AI see content and make smart decisions
2. Precise Over Broad: Make surgical edits to specific elements
3. Clean Over Complex: Remove unnecessary complexity and legacy code
4. Working Over Perfect: Focus on functionality that actually works
5. Document Over Assume: Keep documentation current with implementation

• README-content-editing.md - How the AI analyzes content and converts markdown to PowerPoint XML
• README-development.md - Development roadmap and plans for slide structural operations
• README-how_pptx-works.md - Comprehensive guide to PPTX XML structure, relationships, and editing patterns

Monitoring: GET requests to /health return "PPTX Bot is running" for health checks.