Beamer Academic

Beamer Academic

Generate publication-quality Beamer slides from an academic paper in one automated pipeline.

Pipeline Overview

论文 → 素材提取 → 大纲生成 → [用户确认] → 内容填充 → 编译 → [交互修改循环]

Phase 0: Environment Check & Input Clarification

0.1 Locate Thesis File

Search current directory for thesis in priority order:

1. .tex (LaTeX source — best quality, figures/equations directly reusable)
2. .docx (Word — good for text extraction, figures embedded)
3. .pdf (PDF — acceptable but figure extraction may lose quality)

If multiple candidates found, ask user which one. If none found, prompt: "请把论文文件放到当前目录（支持 .tex / .docx / .pdf）"

Recommend .tex or .docx over PDF: PDF figure extraction can produce low-quality raster images. If user only has PDF, warn:

注意：PDF 提取的图片可能有质量损失。如果你有论文的 Word 或 LaTeX 源文件，建议优先使用。

0.2 Check LaTeX Environment

which xelatex

If missing, recommend user to install (this is a one-time setup, not optional):

你的系统还没有安装 LaTeX 环境。beamer PPT 需要 xelatex 编译，我来帮你装一下：

Then provide the command for user's OS and offer to run it:

Wait for installation to complete, verify with which xelatex, then proceed.

If user explicitly refuses to install (rare), fallback: deliver .tex + .sty + figures, suggest Overleaf.

0.3 Beamer Template (Ask User First)

Ask user about existing school beamer template:

你的学校/课题组有现成的 beamer 模板吗？（比如导师或师兄给过的 .sty 文件）

• 有：请放到当前目录，我直接使用
• 没有：我来帮你找或用内置通用模板

If user has a template: use it directly, skip the built-in theme.

If user has NO template:

1. Ask for school name
2. Search for existing public beamer templates for that school:
   • GitHub search: {school name} beamer template
   • Common sources: Overleaf Gallery, CTAN
3. If found: download and use (notify user which template)
4. If not found: use built-in assets/beamerthemeAcademic.sty with school logo
   • Ask: "你能提供学校 logo 图片吗？（放到当前目录即可）"
   • If no logo available: search online for the school's logo, or proceed without logo

0.4 Configuration

Check for config.yaml in current directory:

• Exists: read and use.
• Missing: ask user for basic info, then generate from template at assets/config.yaml.

Required user info (ask if not in config):

• Institution name and department
• Author name, supervisor, major
• Report type: defense | proposal | conference
• Color preference: blue | red | green | purple | teal
• Time limit (minutes): affects page count and content density
• Language: thesis language vs. PPT language (e.g., 英文论文 → 中文PPT)
• Chapnote preference: "每页要不要显示'对应论文 §X.X'的标注？"（some users find it helpful for committee, others find it cluttered）

0.5 Language Strategy

If thesis language ≠ PPT language, establish rules upfront:

你的论文是英文，PPT 要做中文版吗？ 对于专业术语，我会：

1. 首次出现用"中文（English）"格式
2. 之后统一用中文
3. 公式/变量名保持英文不翻译

这样处理可以吗？

Build a terminology mapping table (stored in materials/terms.md):

| English | 中文 | 首次出现页 |
|---------|------|-----------|
| disparity | 形态多样性 | P4 |
| diversity | 分类丰富度 | P4 |
| Ornstein-Uhlenbeck | OU 过程 | P8 |

Ensure 术语一致性: same concept uses same translation across all pages.

Phase 1: Material Extraction

Create materials/ directory. Extraction strategy depends on input format:

From .tex source (best)

1. Figures: locate \includegraphics paths, copy originals → materials/figures/
2. Tables: extract tabular/table environments → materials/tables/
3. Equations: extract equation/align/$$ blocks → materials/equations.md
4. Structure: parse \chapter/\section hierarchy → materials/structure.md

From .docx

1. Figures: extract embedded images (python-docx or unzip) → materials/figures/
2. Tables: extract table contents → materials/tables/
3. Equations: extract OMML/LaTeX equations → materials/equations.md
4. Structure: parse heading hierarchy → materials/structure.md

From .pdf (fallback)

1. Figures: use pdfimages or read PDF for embedded images → materials/figures/
   • ⚠️ Quality may degrade. Prefer vector (PDF/SVG) extraction when possible.
2. Tables: read and reconstruct → materials/tables/
3. Equations: read and convert to LaTeX → materials/equations.md
4. Structure: parse chapter/section from text → materials/structure.md

1.2 Material Confirmation (In-Conversation)

After extraction, present a figure catalog to user in conversation:

🖼️ 素材库（共提取 12 张图）

#	文件名	来源	内容描述
1	fig_001.png	论文图1	地质年代柱状图
2	fig_002.png	论文图4	PCoA 碎石图
3	fig_003.png	论文图8	形态空间散点图
...

哪些图是你答辩必须展示的？ 可以说：

• "1, 3, 5, 8 必须用"
• "图7 不用了，跟图5 重复"
• "全部都可能用到"

This ensures:

• Key figures won't be missed in layout assignment
• User has a mental model of available materials before the outline phase
• Low-quality or duplicate figures can be flagged early

Phase 2: Brainstorm Outline (Interactive, In-Conversation)

This phase determines the entire presentation structure through conversation with the user. All confirmation happens IN THE CHAT — never ask user to open a file.

2.0 Chapter Structure Principle

Follow the thesis's own structure as the PRIMARY basis for PPT chapters.

• If thesis has explicit chapters (第一章...第四章): map directly to PPT chapters
• If thesis has no chapters (like a conference paper with sections): group related sections into 3-5 logical chapters for the PPT
• Do NOT invent arbitrary chapter divisions that don't match the thesis logic
• The PPT chapter structure should feel natural to someone who has read the thesis

2.1 First Pass: Structure Proposal

After reading the thesis, propose the high-level structure directly in conversation:

📐 PPT 结构提案

根据你的论文结构，我建议这样安排：

总页数：42 页（约 20 分钟答辩）

章	标题	页数	对应论文	说明
一	研究背景与科学问题	8页	第1章	背景+命题+假设+创新点
二	数据构建与描述性分析	7页	第2章	数据来源+特征工程+时序总览
三	演化模式与驱动机制	15页	第3章	H1+H2+H3 三个假设检验
四	结论与展望	5页	第4章	主要结论+局限+展望+成果

你觉得这个结构可以吗？ 可以告诉我：

• "第三章太长了，拆成两章"
• "加一章文献综述"
• "总页数控制在35页以内"

Wait for user to confirm or adjust. Loop until structure is approved.

2.2 Second Pass: Per-Section Detail

Once chapter structure is approved, expand each chapter in conversation (not in a file):

📋 第一章 详细安排（8页）

页码	版式	标题	内容要点
P3	章节分隔	一、研究背景与科学问题	—
P4	左文右图	寒武纪辐射与软体动物的演化窗口	演化窗口介绍 + 地质年代图
P5	纯文段	经典命题：Disparity vs. Diversity	形态多样性与丰富度争论
P6	纯文段	驱动机制之一：外源突变	Sinsk 事件背景
P7	纯文段	驱动机制之二：系统发育约束	OU 过程假说
P8	列表	现有研究的四点不足	4个gap
P9	表格	研究目标与三个核心假设	H1/H2/H3 表格
P10	列表	论文的主要创新	5个创新点

这章的安排可以吗？ 可以说：

• "P5和P6合并成一页"
• "P9加一列'若成立则说明'"
• "ok，继续下一章"

Present one chapter at a time. Wait for approval before showing the next chapter.

2.3 Third Pass: Final Confirmation

After all chapters are individually approved, show a complete summary in conversation:

✅ 最终大纲确认

章	页码范围	页数	版式分布
封面+目录	P1–P2	2	cover, toc
一 研究背景	P3–P10	8	分隔×1, 左文右图×1, 纯文段×3, 列表×1, 表格×1, 列表×1
二 数据构建	P11–P17	7	分隔×1, ...
三 演化机制	P18–P32	15	分隔×1, ...
四 结论展望	P33–P37	5	分隔×1, ...
致谢	P38	1	thanks
合计		38页

确认后我将开始生成 beamer LaTeX 代码。确认吗？

HARD GATE: Do NOT proceed to Phase 3 until user explicitly confirms the final summary.

2.4 Save to outline.md

Only after final confirmation, save the approved structure to outline.md for Phase 3 reference. This file is for the AI's internal use — user has already approved everything in conversation.

Why In-Conversation Instead of File

• Users who don't know markdown can follow along naturally in chat
• Iterative refinement is faster (no "go check that file" round-trip)
• Each chapter gets individual attention

Layout Selection Rules (used in 2.2 per-page assignment)

1. Each chapter start → section-divider
2. Core formula/model definition → formula
3. Multi-row data comparison → table
4. High-information figure (multi-panel, heatmap) → full-image
5. Text-primary with supporting figure → text-left-image-right
6. Figure-primary with interpretation → image-left-text-right
7. Pure concept/background → text-only
8. Chapter-end with clear conclusion → conclusion-box
9. Between chapters → transition
10. Parallel bullet points → list

Rhythm Constraints

• No 3 consecutive pages with same layout
• Each chapter uses at least 3 different layouts
• Total: 35–50 pages (defense), 25–35 (proposal), 15–25 (conference)

Phase 3: Content Generation

1. Copy assets/beamerthemeAcademic.sty to current directory.
2. Generate defense.tex file header. Read references/tex-header.md for template.
3. For each page in outline.md:
   • Load layout skeleton from references/layouts.md (find section by layout id)
   • Fill slots with thesis content + extracted materials
4. Close with \end{document}.

Critical: Section Divider = 满版色 + 圆圈序号

每章开头用 \sectiondivider{1}{章标题}（圆圈里用阿拉伯数字，中文「一」在圆里像减号）。目录只在第 2 页出现一次。

Critical: TOC Page Format

The TOC page (P2) must use Chinese numbering with em-dash subtitles:

\begin{frame}
  \frametitle{汇报提纲}
  \vskip0.3cm
  {\footnotesize
  \begin{tabbing}
  \hspace{0.4cm}\=\hspace{0.6cm}\=\kill
  \textbf{\color{accentcolor}一}\>\textbf{研究动机}\,——\,为什么需要新的序列建模方案\\[8pt]
  \textbf{\color{accentcolor}二}\>\textbf{模型架构}\,——\,Transformer 的核心设计\\[8pt]
  \textbf{\color{accentcolor}三}\>\textbf{实验结果}\,——\,翻译质量与消融分析\\[8pt]
  \textbf{\color{accentcolor}四}\>\textbf{总结与展望}\,——\,贡献与未来方向\\
  \end{tabbing}
  }
\end{frame}