latex-to-ppt - SKILL.md Agent Skill

name: latex-to-ppt description: Generates a rich, structured PPT Markdown (20-30 slides) from a LaTeX thesis or academic paper, ready for use in Gamma, Kimi, or similar AI presentation tools. Use this skill when a user wants to create a presentation from their thesis or research paper.

LaTeX Thesis → PPT Markdown Generator

This skill converts a LaTeX academic thesis into a high-quality, visually structured PPT Markdown (20–30 slides) ready for import into Gamma, Kimi, Beautiful.ai, or any AI presentation generator.

When to Use This Skill

Use this skill when the user wants to:

Create a thesis defense presentation from their .tex file
Generate structured slides from academic research
Produce a high-quality presentation for academic/conference use

Step 1: Read the Thesis

Read the LaTeX thesis file thoroughly. Extract the following:

Title, Author, Supervisor, Affiliation, Date (from cover page)
Abstract (both Chinese and English if present)
Chapter structure (all \chapter{} and \section{} headings)
Core technical content per chapter:
- Problem motivation and significance
- Proposed methods/models and their components
- Key formulas and their intuitive meaning
- Experimental setup (datasets, baselines, metrics)
- Main experimental results (tables, figures, key numbers)
- Ablation study highlights
- Visualization insights (attention heatmaps, etc.)
Conclusion & Future Work
Any numeric results that make strong talking points (e.g., "MSE reduced by 6.7%")

Tip: Pay special attention to \caption{} text in figures and tables — these are often the most slide-ready summaries.

Step 2: Design the Slide Structure

Plan the slides before writing. A good thesis presentation has three layers:

Context Layer (slides 1–5): Why does this problem matter? What's been tried? What's still missing?
Method Layer (slides 6–18): What did we propose? How does it work? Walk through components modularly.
Validation Layer (slides 19–25): What did the experiments show? What does ablation reveal? What do visualizations tell us?
Wrap-up (slides 26–28): Conclusions, future directions, Q&A.

For a two-work thesis, allocate approximately equal depth to each work, with a clear "bridge" slide explaining the progression from Work 1 → Work 2.

Step 3: Write the PPT Markdown

Write a complete Markdown document that can be directly pasted into Gamma, Kimi, or similar tools. Follow these rules:

Structural Rules

Use --- to separate slides
Use # Title for the slide heading
Target 20–30 slides total
Use bullet points (not dense paragraphs) for all content
No more than 5–6 bullets per slide
Each bullet should be one sentence or a short phrase

Content Rules

Lead with numbers: "MSE reduced by 6.7% on Electricity dataset" beats "performance improved"
Name modules clearly: Use the exact names from the paper (e.g., MV-FDE, SAM, TimeBridge)
Contrast baseline vs. proposed: "Standard attention → diluted weights | SAM → focused weights"
Use emoji or icons sparingly to make slides visually scannable
For method slides: describe what → why → how (in that order)

Visualization Guidance

On slides showing figures/tables:

Reference what would appear: [Figure: Architecture diagram showing dual-path flow]
Describe what the visualization proves in 1 sentence
List 2–3 key takeaways as bullets

Slide Template Library

Title Slide:

# [Paper Title]

**Author** | Supervisor: [Name] | [Affiliation]

[Degree type] · [Defense Date]

> [One-sentence thesis statement]

Problem/Motivation Slide:

# Challenge: [Problem Name]

> [One-sentence hook about why this problem matters]

- 📍 **Real impact**: [Application domain example]
- ⚠️ **Core bottleneck**: [What goes wrong technically]
- 📊 **Scale**: [Quantify the difficulty, e.g., "100 variables = 10,000 attention pairs"]
- 🔍 **Root cause**: [Deeper technical reason the problem exists]

Method Overview Slide:

# [Model Name]: Overview

> Design philosophy: "[Slogan or design principle]"

| Stage | Module | Role |
|-------|--------|------|
| Input | [Module A] | [What it does] |
| Encoding | [Module B] | [What it does] |
| Output | [Module C] | [What it does] |

**Key innovation**: [One-sentence differentiator vs. prior work]

Module Deep-Dive Slide:

# [Module Name]

**Goal**: [What problem this module solves]

1. **Step 1**: [Operation] → [Output]
2. **Step 2**: [Operation] → [Output]
3. **Step 3**: [Operation] → [Output]

💡 **Insight**: [Why this design choice is better than the alternative]

Experiment Results Slide:

# Results: [Dataset or Setting Name]

[Figure: Performance comparison table/chart]

**Key findings**:
- 🏆 [Model name] achieves best MSE of **[X.XXX]** on [Dataset]
- 📉 **[X.X]% MSE reduction** vs. [Baseline] on [Dataset]
- 📈 Advantage grows with longer horizon: L=720 shows strongest gains
- 🔑 [Any surprising or noteworthy result]

Ablation Slide:

# Ablation Study

| Variant | [Dataset1] MSE | [Dataset2] MSE |
|---------|----------------|----------------|
| Full model | **X.XXX** | **X.XXX** |
| w/o [Module A] | X.XXX (+X.X%) | X.XXX |
| w/o [Module B] | X.XXX (+X.X%) | X.XXX |
| Baseline | X.XXX | X.XXX |

**Takeaway**: [Module A] + [Module B] = synergistic effect (1+1 > 2)

Visualization/Interpretation Slide:

# Interpretability: [What We're Visualizing]

[Figure: Visualization image]

- 📊 **[Dataset 1]**: shows [Pattern A] → [Interpretation]
- 📊 **[Dataset 2]**: shows [Pattern B] → [Interpretation]
- 🔍 Proves: [What this visualization demonstrates about the model]

Conclusion Slide:

# Summary & Contributions

| Work | Problem Solved | Key Method | Performance |
|------|---------------|------------|-------------|
| [Work 1] | [Problem] | [Method] | [Top metric] |
| [Work 2] | [Problem] | [Method] | [Top metric] |

**Technical Roadmap**: [Metaphor for progression, e.g., "From 1D to 2D modeling"]

> [Closing statement about impact or future direction]

Step 4: Quality Checklist

Before delivering the slides, verify:

Slides 1–3: Hook the audience on why the problem matters
At least 1 slide clearly shows the limitation of prior work (preferably with a diagram or contrast table)
Method section has both an overview slide and per-module slides
At least 2 slides with quantitative results (numbers, percentages)
Ablation results are presented (shows scientific rigor)
At least 1 visualization slide (attention heatmaps, fusion matrices, etc.)
Final slide has a clear conclusion + future work
No slide exceeds 6 bullet points
All module names match the paper exactly
Each slide has a clear "one thing to remember" takeaway

Output Format

Return the complete PPT Markdown as a code block in a .md file, with this header:

<!-- PPT Markdown: [Paper Title] -->
<!-- Slides: [N] | Generated from: [path to .tex file] -->
<!-- Compatible with: Gamma, Kimi, Beautiful.ai, Marp -->

Also provide a brief Slide Index at the top (list of slide numbers and titles) so the user can quickly navigate.