- Home
- Skills
- Research & Analysis
- Paper to Slides
More screenshots
Works with the AI Agents you already use
Paper to Slides
Turn a paper PDF into a minimal group-meeting deck with original figures and a slide-by-slide speaker script.
Free
Paper to Slides
Example session with this skill installed
Turn this paper into a group-meeting deck: papers/attention-is-all-you-need.pdf. I present next Wednesday. Focus on the method, keep it around 9 slides, and write a speaker script for a 15-minute talk.
- Read your context and instructions
- Compiled the paper to slides
Summary
Built a 9-slide deck from the paper PDF and a complete speaker script. The original architecture figure and main results table were cropped from the PDF at 400 dpi; all 9 slides passed the preview check with no overflow or overlap.
Deliverables
| File | Contents |
|---|---|
out.pptx | 9 slides, 16:9, white background with restrained red emphasis |
speaker-notes.md | Slide-by-slide script (15-minute timing) + 5 likely Q&A |
deck.json | Editable copy and figure paths |
figs/fig1.png, figs/fig2.png, figs/table1.png | Original figures/tables cropped from the PDF |
preview/slide-01.png … slide-09.png | Per-slide renders used for visual QA |
Slide outline
| Slide | Title / claim | Visual |
|---|---|---|
| 1 | Attention Is All You Need | Cover |
| 2 | Recurrence serializes sequence modeling | Text |
| 3 | Prior sequence models trade parallelism for locality | Text comparison |
| 4 | The Transformer removes recurrence entirely | Figure 1: architecture |
| 5 | Self-attention connects every position in one step | Figure 2: attention |
| 6 | WMT 2014 En–De and En–Fr measure the claim | Table 1: setup |
| 7 | A new BLEU record at lower training cost | Table 2: main results |
| 8 | Multi-head attention drives the gain | Figure 3: ablations |
| 9 | Attention is sufficient; long-sequence cost remains | Text |
Speaker script excerpt (Slide 4)
"This is the core of the paper: Figure 1. The encoder and decoder are stacks of identical layers, and every layer uses self-attention instead of recurrence. That single choice removes the sequential dependency that limited parallel training. Notice that positional information is added once at the input; the network does not need to process tokens in order. The practical consequence is that the whole sequence can be trained in parallel — which is why the next slides can compare training time, not just accuracy."
Next steps
- Open
out.pptxand confirm the paper title and venue on the cover. - Read
speaker-notes.mdand adjust the timing to your slot. - Spot-check
preview/slide-04.pngandpreview/slide-07.pngagainst the original PDF before presenting.
Connects securely to your tools. The creator never sees your data.
About this skill
Turn a research paper into a compact, claim-driven group-meeting deck — without spending hours on layout.
Paper to Slides reads a paper PDF, follows a fixed 8–12 slide narrative skeleton, crops the paper's own figures and tables losslessly, and produces a clean white/black/red PPTX plus a slide-by-slide speaker script with likely Q&A.
What you get
- A focused .pptx — 8–12 slides, one claim per title, no agenda, logos, or decorative clutter.
- Original figures — cropped at high resolution from the PDF, without redrawing or running heads.
- A speaker script — 80–120 spoken words per slide, plus answers to likely questions.
- Editable source — all copy lives in one
deck.json; change the paper without changing the layout. - Visual checks — per-slide PNG previews so you can catch overflow and collisions before presenting.
Why it works
- A fixed narrative — problem → why prior work fails → method → evidence → limitations.
- Claim titles — every slide says what the audience should remember, not just a section name.
- One layout engine — type sizes and placement stay consistent across every revision.
- PDF-first extraction — handles two-column papers and PDFs without a text layer.
Good fit
Use it for a 10–15 minute group meeting, journal club, literature review, or research update on a systems or CS paper with architecture diagrams or experimental plots. It requires Python 3.9+ and PyMuPDF, python-pptx, and Pillow. Preview rendering uses PowerPoint on Windows or LibreOffice elsewhere. All processing is local.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Research & Analysis.
Requires Python 3.9+ with PyMuPDF, python-pptx, and Pillow. No OCR for scanned PDFs; pages must be rendered and read visually. The visual system is fixed (white/black/red, no custom templates or animations). Best suited to CS/systems papers with figures; equation-heavy theory and surveys are a poor fit. Preview rendering requires PowerPoint or LibreOffice.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 17 days ago
- Free with an Agensi account
- Software
Python 3.9+
- Optional
- Microsoft PowerPointorLibreOffice
PowerPoint or LibreOffice is only for preview rendering.
Permissions
File Scopes
Local-only processing. The skill reads a user-supplied paper PDF and writes figures, deck files, and preview PNGs into the working directory. It may invoke local PowerPoint COM or LibreOffice once to render previews. No network calls, credentials, or external services are used.