references/detailed-guide.md
---
name: slide-automation
description: Generate presentation slides from research content (paper, thesis, markdown) using Python (python-pptx), LaTeX Beamer, Marp, or reveal.js. Use when converting written research into presentation decks, automating figure insertion, or creating reproducible slide templates.
metadata:
tags: ["research", "phd", "presentation", "slides", "powerpoint", "beamer", "marp", "revealjs", "automation", "python-pptx"]
---
# Detailed Slide Automation Guide
> Reference status: Extended examples are tool-specific. Use only the section matching the selected output format and verify the rendered deck.
**From markdown to deck — reproducible, version-controlled presentations.**
This skill automates the creation of research presentations from structured content. It eliminates manual copy-pasting, ensures consistency, and makes slides version-controllable alongside your research.
> "Your slides should be as reproducible as your experiments." — Researcher's Maxim
---
## 1. When to Automate Slides
### Use Cases
| Scenario | Tool | Output |
|---|---|---|
| **Paper → Conference talk** | python-pptx | .pptx with figures auto-inserted |
| **Thesis → Defense slides** | LaTeX Beamer | .pdf with precise typography |
| **Markdown → Quick deck** | Marp | .html or .pdf from markdown |
| **Web-based presentation** | reveal.js | Interactive HTML slides |
| **Recurring reports** | python-pptx | Template-based monthly updates |
### When NOT to Automate
- **One-off, highly designed pitch deck** → Use Keynote/PowerPoint manually
- **Creative storytelling presentation** → Manual design for emotional impact
- **Rapid iteration with designer** → Figma/Sketch collaboration
---
## 2. Tool Comparison
| Tool | Format | Best For | Learning Curve | Collaboration |
|---|---|---|---|---|
| **python-pptx** | .pptx | Data-heavy, figure-rich talks | Low | Git-friendly |
| **LaTeX Beamer** | .pdf | Academic defense, math-heavy | Medium | Git-friendly |
| **Marp** | .md → .pdf/.html | Quick markdown-based decks | Very low | Git-friendly |
| **reveal.js** | .html | Interactive web presentations | Low | Git-friendly |
| **Quarto** | .qmd → multiple | Reproducible research reports | Medium | Git-friendly |
---
## 3. python-pptx: PowerPoint from Python
### 3.1 Basic Structure
```python
from pptx import Presentation
from pptx.util import Inches, Pt
from pptx.dml.color import RgbColor
from pptx.enum.text import PP_ALIGN, MSO_ANCHOR
# Create presentation
prs = Presentation()
prs.slide_width = Inches(13.333) # 16:9
prs.slide_height = Inches(7.5)
# Add title slide
blank_layout = prs.slide_layouts[6] # Blank layout
slide = prs.slides.add_slide(blank_layout)
# Add title
title_box = slide.shapes.add_textbox(Inches(1), Inches(2.5), Inches(11.333), Inches(1.5))
tf = title_box.text_frame
tf.text = "Federated Learning for Medical AI"
p = tf.paragraphs[0]
p.font.size = Pt(44)
p.font.bold = True
p.font.color.rgb = RgbColor(0x1a, 0x1a, 0x1a)
p.alignment = PP_ALIGN.CENTER
# Add subtitle
sub_box = slide.shapes.add_textbox(Inches(1), Inches(4.2), Inches(11.333), Inches(1))
tf = sub_box.text_frame
tf.text = "Privacy-Preserving Collaborative Training"
p = tf.paragraphs[0]
p.font.size = Pt(24)
p.font.color.rgb = RgbColor(0x66, 0x66, 0x66)
p.alignment = PP_ALIGN.CENTER
# Save
prs.save('presentation.pptx')
```
### 3.2 Research Slide Template
```python
from pptx import Presentation
from pptx.util import Inches, Pt, Emu
from pptx.enum.shapes import MSO_SHAPE
from pptx.dml.color import RgbColor
class ResearchSlideDeck:
"""Template for academic/research presentations."""
# Color scheme
PRIMARY = RgbColor(0x1a, 0x5f, 0x9e) # Dark blue
SECONDARY = RgbColor(0x2e, 0x8b, 0x57) # Green
ACCENT = RgbColor(0xe6, 0x8a, 0x00) # Orange
DARK = RgbColor(0x1a, 0x1a, 0x1a) # Near black
LIGHT = RgbColor(0xf5, 0xf5, 0xf5) # Light gray
def __init__(self, title: str, author: str, date: str):
self.prs = Presentation()
self.prs.slide_width = Inches(13.333)
self.prs.slide_height = Inches(7.5)
self.title = title
self.author = author
self.date = date
self._add_title_slide()
def _add_title_slide(self):
"""Add title slide with research branding."""
slide = self.prs.slides.add_slide(self.prs.slide_layouts[6])
# Background accent bar
shape = slide.shapes.add_shape(
MSO_SHAPE.RECTANGLE, Inches(0), Inches(0),
Inches(0.3), Inches(7.5)
)
shape.fill.solid()
shape.fill.fore_color.rgb = self.PRIMARY
shape.line.fill.background()
# Title
title_box = slide.shapes.add_textbox(
Inches(1), Inches(2.5), Inches(11), Inches(1.5)
)
tf = title_box.text_frame
tf.text = self.title
p = tf.paragraphs[0]
p.font.size = Pt(40)
p.font.bold = True
p.font.color.rgb = self.DARK
# Author and date
info_box = slide.shapes.add_textbox(
Inches(1), Inches(4.3), Inches(11), Inches(1)
)
tf = info_box.text_frame
tf.text = f"{self.author}\n{self.date}"
p = tf.paragraphs[0]
p.font.size = Pt(20)
p.font.color.rgb = RgbColor(0x66, 0x66, 0x66)
def add_content_slide(self, title: str, bullets: list[str],
figure_path: str = None):
"""Add a content slide with bullets and optional figure."""
slide = self.prs.slides.add_slide(self.prs.slide_layouts[6])
# Title bar
title_shape = slide.shapes.add_shape(
MSO_SHAPE.RECTANGLE, Inches(0), Inches(0),
Inches(13.333), Inches(1)
)
title_shape.fill.solid()
title_shape.fill.fore_color.rgb = self.PRIMARY
title_shape.line.fill.background()
# Title text
title_box = slide.shapes.add_textbox(
Inches(0.5), Inches(0.2), Inches(12), Inches(0.6)
)
tf = title_box.text_frame
tf.text = title
p = tf.paragraphs[0]
p.font.size = Pt(28)
p.font.bold = True
p.font.color.rgb = RgbColor(0xFF, 0xFF, 0xFF)
# Content area
if figure_path:
# Two-column: bullets left, figure right
# Bullets
content_box = slide.shapes.add_textbox(
Inches(0.5), Inches(1.3), Inches(6), Inches(5.5)
)
tf = content_box.text_frame
tf.word_wrap = True
for i, bullet in enumerate(bullets):
if i == 0:
p = tf.paragraphs[0]
else:
p = tf.add_paragraph()
p.text = f"• {bullet}"
p.font.size = Pt(18)
p.font.color.rgb = self.DARK
p.space_after = Pt(12)
# Figure
slide.shapes.add_picture(
figure_path, Inches(7), Inches(1.3),
width=Inches(5.5)
)
else:
# Full-width bullets
content_box = slide.shapes.add_textbox(
Inches(0.5), Inches(1.3), Inches(12.333), Inches(5.5)
)
tf = content_box.text_frame
tf.word_wrap = True
for i, bullet in enumerate(bullets):
if i == 0:
p = tf.paragraphs[0]
else:
p = tf.add_paragraph()
p.text = f"• {bullet}"
p.font.size = Pt(20)
p.font.color.rgb = self.DARK
p.space_after = Pt(14)
return slide
def add_figure_slide(self, title: str, figure_path: str,
caption: str = None):
"""Add a full-slide figure with caption."""
slide = self.prs.slides.add_slide(self.prs.slide_layouts[6])
# Title
title_box = slide.shapes.add_textbox(
Inches(0.5), Inches(0.3), Inches(12), Inches(0.6)
)
tf = title_box.text_frame
tf.text = title
p = tf.paragraphs[0]
p.font.size = Pt(24)
p.font.bold = True
p.font.color.rgb = self.DARK
# Figure (centered, large)
slide.shapes.add_picture(
figure_path, Inches(1.5), Inches(1.2),
width=Inches(10.333)
)
# Caption
if caption:
cap_box = slide.shapes.add_textbox(
Inches(1), Inches(6.5), Inches(11.333), Inches(0.6)
)
tf = cap_box.text_frame
tf.text = caption
p = tf.paragraphs[0]
p.font.size = Pt(14)
p.font.italic = True
p.font.color.rgb = RgbColor(0x66, 0x66, 0x66)
p.alignment = PP_ALIGN.CENTER
return slide
def save(self, filename: str):
"""Save the presentation."""
self.prs.save(filename)
print(f"Saved: {filename}")
# Usage example
deck = ResearchSlideDeck(
title="Federated Learning for Medical AI",
author="Nguyen Van A",
date="June 2026"
)
deck.add_content_slide(
title="Research Motivation",
bullets=[
"Medical data is siloed across hospitals due to privacy regulations",
"Centralized training requires data sharing — prohibited by HIPAA/GDPR",
"Federated Learning enables collaborative training without data sharing",
"Challenge: Statistical heterogeneity across medical institutions"
]
)
deck.add_figure_slide(
title="Proposed Architecture",
figure_path="figures/fl_architecture.png",
caption="Figure 1: Heterogeneous federated learning with personalized aggregation"
)
deck.save("defense_presentation.pptx")
```
### 3.3 Auto-Generate from Markdown
```python
import re
from pathlib import Path
from pptx import Presentation
from pptx.util import Inches, Pt
def markdown_to_pptx(md_path: str, output_path: str):
"""Convert markdown research outline to PowerPoint."""
with open(md_path) as f:
content = f.read()
prs = Presentation()
prs.slide_width = Inches(13.333)
prs.slide_height = Inches(7.5)
# Parse markdown sections
sections = re.split(r'\n## ', content)
for section in sections[1:]: # Skip title
lines = section.strip().split('\n')
title = lines[0].strip()
# Extract bullets
bullets = []
for line in lines[1:]:
if line.strip().startswith('- ') or line.strip().startswith('* '):
bullets.append(line.strip()[2:])
# Create slide
slide = prs.slides.add_slide(prs.slide_layouts[6])
# Add title
title_box = slide.shapes.add_textbox(
Inches(0.5), Inches(0.5), Inches(12), Inches(0.8)
)
tf = title_box.text_frame
tf.text = title
p = tf.paragraphs[0]
p.font.size = Pt(32)
p.font.bold = True
# Add bullets
if bullets:
content_box = slide.shapes.add_textbox(
Inches(0.5), Inches(1.5), Inches(12), Inches(5)
)
tf = content_box.text_frame
tf.word_wrap = True
for i, bullet in enumerate(bullets[:6]): # Max 6 bullets
if i == 0:
p = tf.paragraphs[0]
else:
p = tf.add_paragraph()
p.text = f"• {bullet}"
p.font.size = Pt(20)
p.space_after = Pt(10)
prs.save(output_path)
print(f"Generated {output_path} with {len(sections)-1} slides")
# Usage
markdown_to_pptx("research_outline.md", "auto_presentation.pptx")
```
---
## 4. LaTeX Beamer: Academic Defense
### 4.1 Basic Template
```latex
\documentclass[aspectratio=169]{beamer}
% Theme
\usetheme{Madrid}
\usecolortheme{seahorse}
\setbeamertemplate{navigation symbols}{}
% Colors
\definecolor{primary}{RGB}{26, 95, 158}
\definecolor{accent}{RGB}{230, 138, 0}
\setbeamercolor{title}{fg=primary}
\setbeamercolor{frametitle}{fg=primary,bg=white}
\setbeamercolor{structure}{fg=primary}
% Information
\title[Federated Learning]{Federated Learning for Medical AI}
\subtitle{Privacy-Preserving Collaborative Training}
\author{Nguyen Van A}
\institute{Hanoi University of Science and Technology}
\date{June 2026}
\begin{document}
% Title slide
\begin{frame}
\titlepage
\end{frame}
% Outline
\begin{frame}{Outline}
\tableofcontents
\end{frame}
% Content
\section{Introduction}
\begin{frame}{Research Motivation}
\begin{itemize}
\item Medical data is siloed across hospitals
\item Privacy regulations prohibit data sharing
\item Federated Learning enables collaborative training
\item Challenge: Statistical heterogeneity
\end{itemize}
\end{frame}
\section{Methodology}
\begin{frame}{Proposed Approach}
\begin{columns}
\column{0.5\textwidth}
\begin{itemize}
\item Personalized aggregation
\item Adaptive learning rates
\item Differential privacy guarantees
\end{itemize}
\column{0.5\textwidth}
\includegraphics[width=\linewidth]{figures/architecture.pdf}
\end{columns}
\end{frame}
\section{Results}
\begin{frame}{Experimental Results}
\begin{table}
\centering
\begin{tabular}{lccc}
\toprule
Method & Accuracy & Rounds & Privacy \\
\midrule
FedAvg & 92.1\% & 134 & $\epsilon=8$ \\
FedProx & 93.0\% & 118 & $\epsilon=8$ \\
\textbf{Ours} & \textbf{94.3\%} & \textbf{87} & $\epsilon=4$ \\
\bottomrule
\end{tabular}
\end{table}
\end{frame}
\section{Conclusion}
\begin{frame}{Conclusion}
\begin{block}{Key Contributions}
\begin{enumerate}
\item Novel aggregation mechanism for heterogeneous data
\item Improved convergence with fewer communication rounds
\item Stronger privacy guarantees
\end{enumerate}
\end{block}
\end{frame}
\begin{frame}
\centering
\Huge Thank You
\vspace{1cm}
\normalsize Questions?
\end{frame}
\end{document}
```
### 4.2 Compile
```bash
# Compile Beamer to PDF
pdflatex defense.tex
pdflatex defense.tex # Run twice for TOC
# Or use latexmk
latexmk -pdf defense.tex
```
---
## 5. Marp: Markdown to Slides
### 5.1 Basic Usage
```markdown
---
marp: true
theme: default
paginate: true
backgroundColor: #fff
---
# Federated Learning for Medical AI
## Privacy-Preserving Collaborative Training
**Nguyen Van A**
Hanoi University of Science and Technology
June 2026
---
## Research Motivation
- Medical data is siloed across hospitals
- Privacy regulations prohibit data sharing
- Federated Learning enables collaborative training
- **Challenge**: Statistical heterogeneity
---
## Proposed Architecture
*Figure 1: Heterogeneous federated learning system*
---
## Results
| Method | Accuracy | Rounds |
|--------|----------|--------|
| FedAvg | 92.1% | 134 |
| FedProx | 93.0% | 118 |
| **Ours** | **94.3%** | **87** |
---
## Conclusion
1. Novel aggregation mechanism
2. Improved convergence
3. Stronger privacy guarantees
---
# Thank You
Questions?
```
### 5.2 Compile
```bash
# Install Marp CLI
npm install -g @marp-team/marp-cli
# Convert to PDF
marp slides.md -o presentation.pdf
# Convert to HTML
marp slides.md -o presentation.html
# Watch mode (auto-rebuild)
marp slides.md --watch
```
---
## 6. reveal.js: Web Presentations
### 6.1 Basic Structure
```html
<!DOCTYPE html>
<html>
<head>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/reveal.js@4.5.0/dist/reveal.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/reveal.js@4.5.0/dist/theme/white.css">
<style>
.reveal h1, .reveal h2 { color: #1a5f9e; }
.reveal .highlight { color: #e68a00; font-weight: bold; }
</style>
</head>
<body>
<div class="reveal">
<div class="slides">
<section>
<h1>Federated Learning for Medical AI</h1>
<p>Privacy-Preserving Collaborative Training</p>
<p><small>Nguyen Van A | HUST | June 2026</small></p>
</section>
<section>
<h2>Research Motivation</h2>
<ul>
<li>Medical data is siloed across hospitals</li>
<li>Privacy regulations prohibit data sharing</li>
<li class="highlight">Federated Learning enables collaborative training</li>
</ul>
</section>
<section>
<h2>Results</h2>
<table>
<tr><th>Method</th><th>Accuracy</th><th>Rounds</th></tr>
<tr><td>FedAvg</td><td>92.1%</td><td>134</td></tr>
<tr><td class="highlight">Ours</td><td class="highlight">94.3%</td><td class="highlight">87</td></tr>
</table>
</section>
<section>
<h1>Thank You</h1>
<p>Questions?</p>
</section>
</div>
</div>
<script src="https://cdn.jsdelivr.net/npm/reveal.js@4.5.0/dist/reveal.js"></script>
<script>Reveal.initialize();</script>
</body>
</html>
```
---
## 7. Integration with AI Figure Generation
```
Research Content (markdown/paper)
↓
[1] Extract figures and concepts
↓
[2] ai-figure-generation → Create conceptual visuals
│ • Architecture diagrams
│ • Process flows
│ • Abstract illustrations
↓
[3] slide-automation → Build slide structure
│ • Insert AI-generated figures
│ • Add data plots (matplotlib)
│ • Format text and layout
↓
[4] Export final deck
│ • .pptx for conference
│ • .pdf for defense
│ • .html for web
↓
Final Presentation
```
### Combined Workflow Example
```python
from ai_figure_generation import generate_diagram_prompt
from slide_automation import ResearchSlideDeck
import matplotlib.pyplot as plt
import numpy as np
# Step 1: Generate AI figure
prompt = generate_diagram_prompt(
concept="federated learning architecture",
style="clean vector diagram",
components=["5 hospitals", "aggregator", "encrypted gradients"]
)
# → Send to DALL-E/Midjourney → Save as 'fl_architecture.png'
# Step 2: Create data plot
fig, ax = plt.subplots(figsize=(8, 5))
rounds = np.arange(1, 101)
ax.plot(rounds, 1 - np.exp(-rounds/30), label='FedAvg', color='gray')
ax.plot(rounds, 1 - np.exp(-rounds/20), label='Ours', color='#1a5f9e')
ax.set_xlabel('Communication Rounds')
ax.set_ylabel('Accuracy')
ax.legend()
ax.set_title('Convergence Comparison')
plt.savefig('convergence_plot.png', dpi=150, bbox_inches='tight')
# Step 3: Build presentation
deck = ResearchSlideDeck("Federated Learning", "Nguyen Van A", "2026")
deck.add_figure_slide("Architecture", "fl_architecture.png")
deck.add_figure_slide("Results", "convergence_plot.png",
"Faster convergence with proposed method")
deck.save("presentation.pptx")
```
---
## 8. Best Practices
### 8.1 Slide Design Rules
| Rule | Why | Implementation |
|---|---|---|
| **1 idea per slide** | Cognitive load | One title = one takeaway |
| **6×6 rule** | Readability | Max 6 bullets, max 6 words each |
| **Figure > Table > Text** | Visual processing | Convert tables to charts when possible |
| **Consistent branding** | Professionalism | Same colors, fonts, layout throughout |
| **Backup slides** | Q&A preparation | 5-10 extra slides after "Thank You" |
### 8.2 Version Control
```bash
# Store slides alongside research
git add slides/
git commit -m "slides: update results with experiment 042"
# Tag versions
git tag v1.0-defense
git tag v1.1-conference
```
### 8.3 Reusable Templates
```python
# Save template for future use
import json
template = {
"colors": {
"primary": "#1a5f9e",
"secondary": "#2e8b57",
"accent": "#e68a00"
},
"fonts": {
"title": "Arial Bold",
"body": "Arial"
},
"layout": {
"title_height": 0.8,
"content_top": 1.5,
"margin": 0.5
}
}
with open("slide_template.json", "w") as f:
json.dump(template, f, indent=2)
```
---
## Integration with Other Skills
| This skill provides | Related skill | For deeper dive |
|---|---|---|
| Slide structure | `defense-prep` | Defense-specific slide organization |
| Figure insertion | `ai-figure-generation` | AI-generated conceptual visuals |
| Data plots | `experiment-tracking` | Publication-ready result figures |
| Academic tone | `technical-english-cs` | Slide text refinement |
| Presentation delivery | `defense-prep` | Q&A preparation, timing |
---
## References
- [python-pptx](https://python-pptx.readthedocs.io/) — Python library for PowerPoint
- [LaTeX Beamer](https://ctan.org/pkg/beamer) — LaTeX presentation class
- [Marp](https://marp.app/) — Markdown presentation ecosystem
- [reveal.js](https://revealjs.com/) — HTML presentation framework
- [Quarto](https://quarto.org/) — Reproducible research publishing
- [ai-figure-generation](../../ai-figure-generation/SKILL.md) — AI-generated figures
SKILL.md
---
name: slide-automation
description: Create or update reproducible presentation decks from papers, theses, reports, Markdown, data, or figures using PowerPoint, Beamer, Marp, reveal.js, or repository tooling. Use when the user needs slide structure, automated deck generation, figure placement, or rendered-deck verification.
metadata:
tags: ["slides", "presentation", "research", "powerpoint", "beamer", "automation"]
---
# Slide Automation
Design for the audience and speaking situation, then automate repeatable layout and data binding. A generated source file is not finished until the rendered deck has been inspected.
## Workflow
1. Establish audience, purpose, duration, venue/screen, output format, template, and accessibility constraints.
2. Extract one claim or question per slide and build a narrative:
- context/problem;
- evidence or method;
- key result;
- interpretation and limits;
- decision or takeaway.
3. Select the tool already used by the project:
- PowerPoint for broad editing and office workflows;
- Beamer for LaTeX-heavy academic material;
- Marp/reveal.js for version-controlled web/Markdown delivery;
- a supplied presentation skill/tool when the environment provides one.
4. Reuse the user's template, theme, fonts, and assets.
5. Generate charts/figures from source data when possible and preserve provenance.
6. Render every slide and inspect montage plus full-size high-risk slides.
7. Fix overflow, clipping, contrast, legibility, missing fonts/assets, animation dependence, and inconsistent references.
## Content Rules
- Use a declarative title that states the slide's message.
- Prefer one visual argument over paragraphs.
- Keep essential labels readable at presentation distance.
- Show units, uncertainty, sample size, and source for quantitative claims.
- Do not stretch, crop, recolor, or simplify scientific figures in a way that changes meaning.
- Separate results from interpretation and limitations.
- Keep speaker notes distinct from visible slide content.
- Add alt text or an accessible equivalent where the format supports it.
For defenses or technical talks, align slide count to actual rehearsal time rather than a fixed slides-per-minute rule.
## Automation Contract
- deterministic input paths and ordering;
- explicit theme/layout constants;
- no machine-specific absolute paths;
- graceful handling of missing assets;
- pinned dependencies when reproducibility matters;
- generated output plus source and build command;
- no secrets or private speaker notes in public artifacts.
## Verification
Check:
- deck opens in the target application;
- every slide renders at intended aspect ratio;
- text and visuals remain inside safe margins;
- equations, Unicode, fonts, and citations survive export;
- charts match source data;
- links/media work offline if required;
- PDF export matches the editable deck;
- no placeholder or example content remains.
## Output
Report artifact paths, source inputs, tool/build command, slide count/story, visual checks performed, and any fonts/assets or manual steps still required.
Read [references/detailed-guide.md](references/detailed-guide.md) only for the selected output technology. For new scientific diagrams or illustrations, use [`ai-figure-generation`](../ai-figure-generation/SKILL.md) and verify scientific accuracy before insertion.