---

name: scientific-slides description: Build and revise computer science research slides for conference talks, paper presentations, defenses, group meetings, demos, and project updates. Use when turning a paper, benchmark, experiment, system design, or research plan into clear slides with strong structure, visual hierarchy, and timing discipline.

Scientific Slides

Use this skill for CS research presentations. Optimize for technical clarity, narrative flow, and audience comprehension under time pressure.

Start With Talk Constraints

Clarify or infer:

• talk type: conference talk, paper reading, defense, seminar, lab update, demo, tutorial
• audience: subfield experts, broad CS audience, students, industry engineers, reviewers
• duration and Q&A budget
• medium: Beamer, PowerPoint, Markdown-to-slides, or another deck workflow
• whether the deck supports a paper, a project demo, a proposal, or a benchmark report

Read references/talk_types_guide.md or references/presentation_structure.md when structure needs help.

Narrative for CS Talks

Most strong computing talks follow this sequence:

1. problem and why it matters
2. why the problem is hard
3. core idea, architecture, or system design
4. how the method / system / study works
5. evaluation setup
6. key results and what they mean
7. limitations, takeaways, and next steps

Do not mirror the paper section-by-section if that hurts comprehension.

Slide-Level Rules

• Keep one main idea per slide.
• Prefer diagrams, tables, and concise bullets over paragraph text.
• Make architecture, pipeline, benchmark setup, or experimental protocol visually explicit.
• Highlight the comparison that matters; do not dump full result tables when a distilled view is enough.
• Use animations sparingly and only when they reveal structure or sequence.

For result slides, answer three questions fast:

• what is being compared
• what the main result is
• why the result matters

Visual Content That Matters in CS

Prioritize visuals that clarify technical structure:

• model or system architecture
• data / control flow
• training or inference pipeline
• benchmark composition and evaluation protocol
• ablation summaries
• failure analysis or qualitative examples
• latency / memory / throughput tradeoffs

Read references/data_visualization_slides.md and references/slide_design_principles.md when revising cluttered result slides.

Use Local Assets and Scripts

For LaTeX/Beamer work:

• assets/beamer_template_conference.tex
• assets/beamer_template_defense.tex
• assets/beamer_template_seminar.tex
• references/beamer_guide.md

For PowerPoint-style work:

• assets/powerpoint_design_guide.md
• assets/timing_guidelines.md

For production and review:

• scripts/slides_to_pdf.py
• scripts/validate_presentation.py
• scripts/pdf_to_images.py

Only use scripts/generate_slide_image.py or scripts/generate_slide_image_ai.py if the user explicitly wants synthetic illustration support. Do not default to decorative AI images.

Review Checklist

Before finalizing a deck, check:

• the first 3 slides establish the problem and context fast
• every result slide has an interpretable takeaway
• fonts and figure labels remain readable at presentation distance
• timing is realistic for the allotted slot
• backup slides exist for details that may come up in Q&A

Default Deliverables

When the user asks for slide help without specifying format, return:

• a talk outline
• slide-by-slide message plan
• recommendations for figures and tables
• timing notes
• revision priorities for clarity