PowerPoint Generation from Conference Photos

A sophisticated multi-agent system that automatically transforms photos from technical conferences and exhibitions into professional PowerPoint presentations.

Overview

This skill orchestrates four specialized agents to process conference photos through a complete pipeline: image analysis with OCR and chart reconstruction, content synthesis with storytelling, presentation design, and quality assurance. The system creates structured workspaces and generates publication-ready presentations.

Agent Architecture

The system uses four specialized agents coordinated by `presentation-orchestrator`:

1. **image-analyzer**: Processes photos, extracts text via OCR, generates Markdown descriptions, reconstructs charts/tables

2. **content-synthesizer**: Reads analysis output, organizes content into logical outlines, adds technical insights

3. **presentation-designer**: Creates PowerPoint presentations using `python-pptx` based on outlines and materials

4. **quality-assurance-editor**: Reviews final presentation for content accuracy and format quality

Instructions

Step 1: Initialize Project Workspace

When receiving a request, create this workspace structure:

```

project_exhibition_summary/

├── README.md # Project documentation

├── 00_inputs/

│ └── raw_photos/ # User's original photos (read-only)

├── 01_processing_outputs/

│ ├── extracted_markdown/ # MD files from image-analyzer

│ └── reconstructed_figures/ # Reconstructed charts/tables

├── 02_src/

│ ├── __init__.py

│ ├── content_synthesis.py # Content analysis functions

│ └── pptx_generator.py # PPT generation functions

└── 03_deliverables/

├── presentation_outline.md # Final outline (optional)

└── exhibition_summary_final.pptx # Final presentation

```

**Note**: The folder `00_inputs/raw_inputs` already exists with user photos. Only create the other directories.

Step 2: Phase One - Image Analysis

Launch `image-analyzer` agent with task:

- Extract text content using Claude Sonnet's vision capabilities

- Generate structured Markdown file in `01_processing_outputs/extracted_markdown/`

- If photo contains charts/tables, reconstruct them using `matplotlib`/`pandas`/`seaborn` and save to `01_processing_outputs/reconstructed_figures/`

**Technical Guidelines for image-analyzer**:

Step 3: Phase Two - Content Synthesis

Launch `content-synthesizer` agent with task:

- Title slide

- Section organization

- Key points per slide

- Where to place photos and reconstructed charts

Step 4: Phase Three - Presentation Generation

Launch `presentation-designer` agent with task:

- Add title and content text

- Embed original photos from `00_inputs/raw_photos/`

- Embed reconstructed charts from `01_processing_outputs/reconstructed_figures/`

- Apply professional formatting and layout

Step 5: Phase Four - Quality Assurance

Launch `quality-assurance-editor` agent with task:

- Content accuracy and technical correctness

- Spelling and grammar

- Slide layout and visual consistency

- Chart readability

- Appropriate image placement

Step 6: Deliver Results

As `presentation-orchestrator`:

- Number of photos processed

- Number of slides created

- Key topics covered

- Agent execution summary

- Path to final presentation

- Summary of content

- Instructions for accessing deliverables

Pipeline Flow

```

User → presentation-orchestrator → image-analyzer → content-synthesizer → presentation-designer → quality-assurance-editor → User

↓ ↓ ↓ ↓ ↓

Initialize Generate MDs Create outline Generate PPT Review & finalize

workspace & charts

```

Python Environment

- `python-pptx`: PowerPoint generation

- `matplotlib`, `pandas`, `seaborn`: Chart reconstruction

- `graphviz`: Flowchart generation (optional)

Important Constraints

1. **Skip unclear photos**: Do not analyze or include photos that are difficult to recognize

2. **Avoid duplication**: Skip photos with extremely similar content

3. **Technical tone**: Write in professional tech article style, avoid emotional language

4. **Context assumption**: Photos are from AI-related technical conferences, seminars, or vendor demonstrations

5. **Existing folder**: `00_inputs/raw_inputs` already exists with user photos—do not recreate it

Example Usage

**User request**: "Create a presentation from the photos in my conference folder"

**Orchestrator response**:

1. Initialize workspace structure under `project_exhibition_summary/`

2. Launch image-analyzer to process 47 photos → generate 42 MD files + 18 reconstructed charts

3. Launch content-synthesizer → create 6-section outline with 24 slides

4. Launch presentation-designer → generate draft PPT with embedded photos and charts

5. Launch quality-assurance-editor → review and finalize presentation

6. Deliver: `03_deliverables/exhibition_summary_final.pptx` (24 slides covering AI product demos, technical architectures, and key insights)