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

When to Use

• You have one or more research paper PDFs and need a conference-style poster draft with standard sections (Intro/Methods/Results/Conclusions).
• You want to automatically extract title/authors/abstract/sections from a PDF and restructure them into concise poster bullet points.
• You need a LaTeX poster scaffold using beamerposter, tikzposter, or baposter, with standard poster sizes (A0/A1/36×48").
• You must ensure posters are visual-first (at least 2–3 generated figures) and pass an automated quality gate before delivery.
• You want a browser-viewable final output (poster.rendered.html) rather than a compiled PDF (PDF compilation is optional).

Key Features

• End-to-end pipeline: PDF → metadata extraction → content structuring → figure generation → LaTeX poster assembly → HTML base conversion → agent rendering → final HTML.
• Scripted PDF processing:
  • Convert PDF pages to images for reference.
  • Extract metadata (title, authors, abstract, section structure).
  • Structure content into poster-ready bullet points with word limits.
• Mandatory figure workflow:
  • Generate at least 2–3 figures (schematics/flowcharts/mechanisms/comparison charts).
  • Insert figures into LaTeX templates automatically or via config.
• Template support:
  • assets/templates/beamerposter-template.tex
  • assets/templates/tikzposter-template.tex
  • assets/templates/baposter-template.tex
• HTML-first delivery policy:
  • Final deliverables: poster.rendered.html + figures/
  • Intermediate artifacts are temporary and must not be returned.
• Quality control:
  • Automated checks for HTML/LaTeX outputs.
  • Manual checklist reference: references/poster_quality_checklist.md
• Design guidance:
  • Poster design principles: references/design_principles.md

Dependencies

Python (recommended)

• Python >=3.8
• pypdf (version not pinned)
• pdfplumber (version not pinned)
• pdf2image (version not pinned)
• pytesseract (version not pinned; required for OCR on scanned PDFs)
• pandas (version not pinned; optional for table handling)
• Pillow (version not pinned)
• matplotlib (version not pinned)

Install (example):

pip install pypdf pdfplumber pdf2image pytesseract pandas pillow matplotlib

System tools

• tesseract-ocr (for OCR; required only for scanned PDFs)
• Poppler utilities (commonly required by pdf2image, e.g., pdftoppm)

LaTeX (optional, only if compiling PDF)

• TeX Live / MiKTeX / MacTeX
• TeX packages (via tlmgr, names may vary by distribution):
  • beamerposter
  • tikzposter
  • baposter
  • qrcode
  • xcolor
  • tcolorbox
  • subcaption
  • graphics

Example:

tlmgr install beamerposter tikzposter baposter qrcode xcolor tcolorbox subcaption graphics

HTML conversion

• pandoc (required for --html-only base HTML generation)
• pdf2htmlEX (optional; only if doing PDF → HTML rendering)

Example Usage

1) Run the end-to-end pipeline (recommended)

Creates a timestamped directory under output/ and supports HTML-only mode.

python scripts/run_poster_pipeline.py paper.pdf --html-only

Expected final structure:

output/YYYYMMDD_HHMMSS/
├── poster.rendered.html
└── figures/
    ├── mechanism.png
    ├── process_flowchart.png
    └── comparison_chart.png

2) Run stages manually (fully reproducible)

Stage A — Extract metadata and structure content (temporary artifacts)

python scripts/extract_metadata.py paper.pdf metadata.json
python scripts/structure_content.py paper.pdf --json poster_content.json --latex content.tex --max-words 800
python scripts/convert_pdf_to_images.py paper.pdf paper_images/ --dpi 600

Stage B — Generate figures (mandatory)

At least 2–3 figures are required.

python scripts/generate_figures.py schematic "Cell signaling pathway" mechanism.png
python scripts/generate_figures.py flowchart "Step 1;Step 2;Step 3" process_flowchart.png
python scripts/generate_figures.py comparison '{"Control":[65],"Our Method":[87]}' comparison_chart.png

Or generate from a config:

python scripts/generate_figures.py config figures_config.json

Stage C — Create poster LaTeX from a template (temporary artifact)

cp assets/templates/beamerposter-template.tex poster.tex
# (Edit poster.tex: title/authors/institute + paste structured content)

Stage D — Insert figures into LaTeX (temporary artifact)

python scripts/insert_figures.py poster.tex --all
# or:
python scripts/insert_figures.py poster.tex --config figures_config.json

Stage E — Convert to base HTML and render final HTML (final artifact)

python scripts/convert_poster.py poster.tex --html-only
# Agent step: read poster.html, validate <img> paths, inject CSS/layout, output poster.rendered.html

Stage F — Quality control (mandatory)

python scripts/check_poster_quality.py poster.rendered.html

Implementation Details

Pipeline and file policy

All generated files must be placed under output/ (preferably a timestamped subdirectory). Only the following are considered final deliverables:

• poster.rendered.html
• figures/ directory (PNG figures)

All other files are intermediate/temporary and must not be returned to the user, including (non-exhaustive):

• poster.tex, poster.html
• metadata.json, poster_content.json, figures_config.json, figures.json
• LaTeX auxiliary files (.aux, .log, .out, etc.)

Conceptual flow:

PDF → metadata extraction (temp) → content structuring (temp) → figure generation (final figures/)
→ LaTeX poster (temp) → base HTML (temp) → agent rendering → poster.rendered.html (final)

PDF extraction approach

• Text and tables are extracted via pdfplumber.
• For scanned PDFs, OCR is performed by converting pages to images (pdf2image) and running pytesseract.

Typical extraction targets:

• Title/authors (first-page patterns)
• Abstract (from “Abstract” header to next section)
• Section blocks (Introduction/Methods/Results/Conclusion)
• Tables (converted to summaries for poster bullets)

Content structuring rules

• Convert paragraphs to bullet points suitable for poster blocks.
• Enforce a poster-friendly word budget (commonly 600–800 words total).
• Prefer 3–6 key visuals; summarize tables into a small set of metrics.

Figure requirements (mandatory)

• Every poster must include at least 2–3 generated figures.
• Target 40–50% of poster area as visual content.
• Figures should be ≥300 DPI, with clear labels and concise captions.

Supported figure categories:

• Schematics (systems, pathways, conceptual diagrams)
• Flowcharts (procedures, pipelines, algorithms)
• Mechanism diagrams (biological/chemical processes)
• Comparison charts (bar charts, benchmarks)

LaTeX package selection

Use one of the supported poster packages depending on style needs:

• beamerposter (traditional academic)
• tikzposter (modern, colorful)
• baposter (multi-column layouts)

Templates are provided in assets/templates/.

HTML rendering requirements (agent step)

After pandoc produces poster.html, the renderer must:

1. Verify all <img> references exist and paths resolve to figures/.
2. Apply poster-grade CSS (grid columns, typography, spacing, captions).
3. Ensure accessibility: WCAG AA contrast ≥ 4.5:1.
4. Output poster.rendered.html as the only final HTML artifact.

Minimum layout expectations:

• 2–3 column grid on desktop; single column on narrow screens.
• Clear hierarchy: title/authors/institution, section headers, bullet lists, figure blocks with captions.

Quality control requirements (mandatory)

Run scripts/check_poster_quality.py on the final output:

• Validate that images exist and render correctly.
• Confirm layout integrity (columns/blocks).
• Ensure readability (font sizes, spacing).
• Confirm contrast compliance (≥ 4.5:1).
• Confirm figure count (≥ 2–3) and that no placeholder text remains.

References:

• Design principles: references/design_principles.md
• Manual checklist: references/poster_quality_checklist.md