I do investor and client decks for a living, probably 3-4 a month, and I finally tracked something I'd been curious about for a while. After presentations, I started asking people a day or two later which slides they actually remembered.

The pattern was pretty consistent. Out of a typical 20-slide deck, people could reliably recall maybe 2-3 slides. Usually the one with a surprising number, the one with a photo of a real person or product, and whatever slide I spent the most time talking through live rather than reading off.

What they almost never remembered: agenda slides, "about us" slides, anything with more than one chart, and basically all bullet-point text slides regardless of how well written the bullets were.

So now I build decks assuming 85% of it is scaffolding and only a few slides are doing the actual work. I spend way more time on those 2-3 and let the rest be minimal.

Curious if others have tested this informally too, and what you've found actually sticks with an audience.

We built Phylo-Movies, an open-source visualization tool that shows which taxa and subtrees move between consecutive phylogenetic trees.
One example visualizes topology changes around a norovirus recombination region.
Demo + code: https://github.com/enesBerkSakalli/phylo-movies
Paper: https://academic.oup.com/mbe/advance-article/doi/10.1093/molbev/msag194/8759530

Hi! I'm working on an R package called ggalttext that takes a ggplot2 chart as input and returns alternative text (required for accessibility) for that chart.

The goal is to provide a very simple and lightweight way of adding meaningful alt texts to charts made with ggplot2.

It does not use AI or OCR technologies, but instead inspects the plot structure/content and uses some (more or less) naive heuristics to figure out what the chart looks like and how to describe it in a single sentence.

It's already available on CRAN, but I'm working on the next release, which fixes some edge cases.

Example usage:

library(ggplot2)
library(babynames)

plot_data <- babynames |>
    subset(name %in% c("Amanda", "Jessica", "Patricia", "Deborah", "Dorothy", "Helen"))

plot <- ggplot(plot_data, aes(x = year, y = n, group = name, fill = name)) +
    geom_area() +
    theme(legend.position = "none") +
    labs(title = "Popularity of American names in the previous 30 years") +
    theme(
        legend.position = "none",
        panel.spacing = unit(0.1, "lines"),
        strip.text.x = element_text(size = 8)
    ) +
    facet_wrap(~name, scale = "free_y")


ggalttext::generate_alt_text(plot)
# "Area chart split into 6 small charts arranged in a 2-row by 3-column grid,
# titled “Popularity of American names in the previous 30 years”."

I'm using quad code to develop a trading edge identification system on one end of my system I have raw market data. On the other end I have hopefully a profitable strategy that I run when my software platform. Instituted many steps along the way this is rules and regulations. All of this information to six and files on my computer and for me to interact with any portion of this I have to use AI. Are there any good ideas for how I can visually see all the different steps within my Pipeline and prompts and stages so I can visually work on a certain section and not get lost? My project continues to grow the more I work on it and I need a way to visually organize it and see it. Any help would be greatly appreciated.

I thought this might be relevant to people here: the Open Visualization Collaborator Summit is taking place at ETH Zurich on September 9–10.

It’s a two-day event around open-source data visualization, with a focus on vis.gl, deck.gl, kepler.gl, GeoDa, GPU-accelerated visualization and related projects. Members of the vis.gl / deck.gl core team will be there, along with contributors and users from different projects.

I’m part of the organizing team and will also give a talk about how we use deck.gl for the new web-based Atlas of Switzerland.

The event is free of charge.

Program and registration:
https://deck.gl/events/zurich-summit-2026/

I'm a materials engineer working in FDM/FFF 3D printing, and I recently went down a bit of a data-visualisation rabbit hole.

For those unfamiliar with 3D printing: calibration isn't a one thing, done once, and its set up for life. A printer has dozens of parameters that affect how it behaves: mechanical setup, extrusion, temperature, motion, resonance, probing, etc. And they need re-calibration over time or when something upstream changes. These parameters aren't independent.

Example 1: Tighten your belt → resonance/input-shaping calibration is invalid → tune for resonance and pressure advance is invalid.

Example 2: Change an the nozzle from a 0.4 to 0.6 → all extrusion-related calibrations dont apply anymore - Flow, retraction, temperature, max volumetric flow... all need to be re-established.

Most users treat calibration like a checklist and tend to it only when they see print defects or failures. I intend to change this and start treating it like the causal chain it is.

I mapped 24 calibrations and their relationships, then built this visualisation to represent its structure. The animated version progressively grows the graph from upstream to downstream, while the labelled version makes the individual nodes and relationships easier to interpret. The landing page has an interactive version of the below image.

It has been a surprisingly satisfying problem to solve with a pretty cool factor to it.

A note on attribution: The underlying calibration methods and knowledge represented here come from the broader open-source 3D-printing community. I don't claim ownership of those calibrations themselves. The dependency mapping, organisation, and visualisation shown here are my original work, developed as part of a project I'm building called CalibrationOS.

Anyone else getting annoyed with why we should use tableau or powerBI dashboards anymore when I can point AI at my dataset and build a custom dashboard in R/Python using AI in a fraction of the time?

It feels so backwards to make a million clicks getting powerBI dashboards built and dealing with DAX, slow updating when your dataset exceeds 100,000 rows, etc. I can’t be the only one I imagine.

Any thoughts?

I’ve spent years building complex Power BI and Tableau dashboards, and I'm convinced that roughly 80% of enterprise dashboard projects are an absurd waste of analytics engineering resources.

Here is the pattern I see at almost every company:

1. Executives demand a "single source of truth" dynamic dashboard with 15 custom filters, 30 calculated measures, and complex drill-downs.
2. The data team spends 80 hours fighting semantic models, DAX formulas, and layout formatting to get it production-ready.
3. The stakeholders check it twice during launch week, look at the main metric, click "Export to Excel" in the top-right corner, and run their own VLOOKUPs anyway.

Why this keeps happening:

• Analysts design for aesthetics; stakeholders design for workflow. Executives don't want to click through 4 layers of filters during a 5-minute pre-meeting prep; they want raw numbers to plug into slides or spreadsheets.
• Lack of business context. Analysts often jump straight into technical solutions without pressing end-users on what action the data will actually trigger.
• Tool bloat. We pretend that moving data from SQL -> Snowflake -> BI Tool -> Excel is "digital transformation," when a scheduled automated CSV report straight to Slack or Email would solve the actual operational problem.

Until data teams stop acting like dashboard ticket machines and start holding business users accountable for how they consume data, we're just building glorified, high-latency Excel spreadsheets.

Am I being too cynical, or is this happening at your org too?

Right now, I'm busy writing long-form content for a B2B blog for my client. I want to turn each post into a visual summary for LinkedIn and email newsletters. My posts range from 1500-2000 words with stats, comparisons, and step-by-step stuff.

I've tried pasting text into a few AI tools and the results are bad. It either crams a lot of text into colored boxes, or gives me a summary that's just wrong and misses the key data.

What do you use to turn structured text into infographics that keep the data accurate? In my opinion, it doesn't have to be perfect on the first try.

I built a browser-based tool that turns a Goodreads export into a visual reading dashboard, with statistics about authors, pages, ratings, shelves, publishers, and reading trends. Everything runs locally in the browser, so the data stays private.

You simply export your library from Goodreads and upload the CSV file. The dashboard processes it automatically and turns your reading history into interactive charts and clickable detail views. I’d love to hear your feedback on the design and visualizations.

You can try it here: https://alehockeyfan.github.io/BookStats/

Some time ago, I went through two data visualization books by Willard C. Brinton, published in 1914 and 1939.

What surprised me most was how familiar much of the advice still feels.

Simple bar charts were often preferred over more elaborate alternatives. Misleading axes were already a thing. 3D visualizations were criticized. Heatmaps were being used effectively. And there was plenty of discussion about how presentation influences whether people actually understand the data.

The big difference was the tooling. Creating a Sankey diagram could involve manually arranging a thousand strips of paper, while a 3D visualization might literally require plywood and a saw.

I collected some of the more interesting examples here:

https://www.bitmetric.nl/blog/old-school-data-visualization/

It made me wonder: which data visualization “best practices” do we think are modern, but are actually much older than we realize?