Generate PowerPoint slides with AI agents. Feed a research paper, thesis chapter, study report, or any structured text to an AI agent — get back a polished .pptx with speaker notes, consistent styling, and editable source code you can diff and version-control.
deckgen is an agent skill that teaches an AI to produce
pptxgenjs-based presentations. You describe what you want — or hand the agent
a document — and it generates a .js file that builds the .pptx on node
run. Every deck uses the same visual system (navy/teal/gold palette, Cambria +
Calibri fonts, layout constants) so slides stay consistent without manual
formatting.
"Turn this paper into a 10-minute presentation.
Include speaker notes for every slide."
The agent reads your document, extracts the key contributions, and produces a
.js file with slides for: title, research question, methods, results (tables +
stat cards), key findings (two-panel comparisons), discussion, implications, and
closing. Each slide gets speaker notes that explain the content in natural
language — ready for rehearsal.
This works with:
- Research papers — extract methods, results, and contribution statements
- Thesis chapters — compress a literature review or results chapter into a defence-ready deck
- Study reports — pull stat tables, findings, and recommendations into a client-facing deck
- Literature reviews — compare findings across studies in two-panel layouts
- Grant applications — structure aims, significance, and approach slides
Every slide generated by the skill includes speaker notes — natural-language talking points that walk through the slide content. The agent writes these from your source material, so they're not generic placeholders but actual presentation script you can rehearse with.
Once the agent generates a deck, you can refine it by prompting:
"Make the results slide a table instead of stat cards."
"Add a slide comparing treatment vs control groups."
"Shorten the speaker notes — I speak faster than this."
"Change the colour scheme to match my university brand."
"Add the effect sizes from Table 3 to the key findings slide."
The agent edits the .js file and re-runs the build. You keep full control of
the source.
The template supports embedding images via pptxgenjs image methods. Place
your .png or .jpg files in assets/ and the agent will reference them with
s.addImage(). When no image file is available, the template uses a clean
placeholder with a caption. Contributions welcome for automatic figure
extraction and responsive sizing.
npm install
node build_deck.js # generates Presentation.pptxThe template produces 11 slides with placeholder content demonstrating every layout type.
Navy background with gold accent, topic tags, and author info:
Three cards across with large numbers and coloured top bars — ideal for sample size, follow-up duration, and key metrics:
Navy-header tables on the left, grey takeaway boxes on the right — for showing effect sizes, p-values, and summarising what they mean:
Side-by-side panels with coloured headers — for contrasting two dimensions, subgroups, or theoretical perspectives:
Full-width figure area with placeholder and caption — embed real charts, plots,
or diagrams via s.addImage():
Navy background with centred title and gold accent:
All positions are defined in a LAYOUT constants object, so changing one value
updates every slide that uses it. The template includes four reusable helpers:
addKeyBox— grey card with gold bar, title, and bullet list (warns on overflow)addPanelBox— coloured header bar with content belowaddAccentPanel— left accent bar with section title and bulletsaddResultTable— consistent table layout with navy headers
| File | Purpose |
|---|---|
build_deck.js |
Template with 11 slides, placeholder content, and speaker notes — edit this or let the agent rewrite it |
Presentation.pptx |
Pre-built output from build_deck.js — open directly or regenerate with node build_deck.js |
SKILL.md |
Agent skill: palette, fonts, helpers, slide patterns, anti-patterns |
package.json |
pptxgenjs dependency |
package-lock.json |
Locked dependency versions for reproducible builds |
assets/ |
Preview screenshots and composite overview image |
Vanilla Node.js + pptxgenjs. No build step, no transpilation, no frameworks. The agent skill is a single Markdown file — works with any AI tool that reads files.






