PDF Ppt

Source: https://github.com/aipoch/medical-research-skills

When to Use

• Converting a research paper PDF into a literature-report PPTX with a standard academic slide structure (title → overview → results → conclusion).
• When the user requires figure-by-figure interpretation (including subpanels) and strict 1:1 mapping between each figure/table and slide content.
• When you must work offline-first (no external API calls) to extract text, figures, legends, and tables from a PDF.
• When the PPT must match a provided layout script (positions, aspect ratios, text boxes) and keep consistent styling across the deck.
• When the output narration must be professional Chinese, concise, and technically accurate.

Key Features

• Offline-first PDF parsing to extract:
  • Title, journal, publication date, section text
  • Figure images and legends grouped by figure number
  • Tables and (optionally) graphical abstract
• Full-paper summary in professional Chinese (background + key findings).
• Results interpretation aligned per figure/subfigure, avoiding vague references.
• PPTX generation with:
  • Fixed slide structure (Title / Overview / Figure slides / Final summary)
  • Layout constraints matching a v5-style deck (coordinates, font sizes)
  • Consistent panel-label styling and per-deck label color
  • Per-figure slide “Summary:” one-liner
• Integration guidance for PPTX editing via D:\\SKILL\\project\\PPTX\\SKILL.md.

Dependencies

• Local offline parsing workflow: references/offline-parsing.md (version: N/A, repository-local)
• Figure/table interpretation rules: references/figure-interpretation.md (version: N/A, repository-local)
• PPTX creation/editing skill: D:\\SKILL\\project\\PPTX\\SKILL.md (version: N/A, local path)
• Optional local mapping file: figure_titles_zh.txt (version: N/A, user/project-provided)

Example Usage

Input (user request)

Please convert this paper PDF into a literature-report PPT.
Requirements: interpret each figure, keep figure-to-text alignment, and follow the v5 layout rules.

Expected Workflow Output (offline-first → structured → PPTX)

1. Offline parse the PDF (no external APIs), following references/offline-parsing.md, and produce a local structured package:

   • Cleaned metadata text: title, journal/date, headings
   • Figure legends grouped by figure number
   • Extracted images named:
     • Figure_1.jpg, Figure_2.jpg, ...
   • Graphical abstract image (if present)
   • Draft explication.json aligned to figure order

2. Write Chinese content:

   • Full-paper summary (background + key findings)
   • Figure-by-figure Results interpretation (purpose → method → object → outcome)
   • Concise critical appraisal (core contributions + internal limitations)

3. Generate PPTX with the required slide structure and layout constraints:

   • Slide 1: Title + metadata (journal/date/presenter/report date)
   • Slide 2: Overview (two text boxes; graphical abstract if present)
   • Middle slides: one slide per figure/table in original order
   • Final slide: overall summary + critical appraisal

Minimal runnable “deck plan” (content-to-slide mapping)

{
  "deck_style": {
    "language": "zh-CN",
    "panel_label_color": "#065A82",
    "transition_preset": "fade",
    "entrance_preset": "appear"
  },
  "slides": [
    {
      "type": "title",
      "title": "<Paper Title>",
      "meta": {
        "journal": "<Journal>",
        "publication_date": "<YYYY-MM-DD>",
        "presenter": "XXX",
        "report_date": "<YYYY-MM-DD>"
      }
    },
    {
      "type": "overview",
      "background": "<Chinese background summary>",
      "findings": "<Chinese key findings summary>",
      "graphical_abstract": "Graphical_Abstract.jpg"
    },
    {
      "type": "figure",
      "figure_id": "Figure_1",
      "title": "Figure 1: <Chinese figure title if available>",
      "image": "Figure_1.jpg",
      "body_zh": [
        "**A** ...",
        "**B** ...",
        "**Summary:** ..."
      ]
    },
    {
      "type": "final_summary",
      "core_highlights": ["..."],
      "internal_limitations": ["..."]
    }
  ]
}

Implementation Details

1) Offline-first PDF collection (must not call external APIs)

Follow references/offline-parsing.md and produce a local structured output containing:

• Cleaned text fields:
  • title, journal, publication_date
  • section headings and key paragraphs (abstract/intro/results)
• Figures:
  • images exported and named Figure_1.jpg, Figure_2.jpg, ...
  • legends grouped by figure number
• Tables:
  • extracted table content and captions (as text or images, depending on the workflow)
• Graphical abstract:
  • capture image and approximate placement if present
• Draft explication.json:
  • ordered by figure appearance
  • each entry links legend → image filename → planned slide text blocks

Ignore supplementary items (e.g., Figure S1) unless explicitly requested.