Andrew Holz

Isabel Z is killing it giving a talk on data tools as the keynote speaker at pydata Boston! Let’s go!

I put together two roundups from posit::conf(2025):

• Shiny talks + full workshop materials
shiny.posit.co/blog/posts/c...

• Quarto talks playlist and summaries
quarto.org/docs/blog/po...

When plotting, LLMs see what they expect to see - Posit Data science agents need to accurately read plots even when the content contradicts their expectations. Our testing shows today's LLMs still struggle here.

Introducing bluffbench, a new tool to evaluate how well LLMs actually see data plots.

When we trick LLMs with secret #RStats transformations, they can miss the visual contradiction.

bluffbench helps us measure this "blind spot" in AI coding agents. Learn more: posit.co/blog/introdu...

PyData Boston 2025

I am looking forward to PyData Boston in a few weeks. If anyone is going and wants to meet up please let me know. I will be sitting at the Posit table quite a bit of the time. I went to the last few PyData NYCs so it will be interesting to see how it differs. pydata.org/boston2025

Python Open-Source Developer Posit is hiring a Python open-source developer to create more data analysis tools.

We're hiring an open-source #python developer focused on modeling APIs!

tidyverse.org/blog/2025/11...

#numpy #scipy #scikitlearn

How long before I can get a credit card with a reward program paying out in Anthropic or OpenAI token credits?

Shiny - Build Your First LLM App with Shiny for Python or R Shiny is a package that makes it easy to create interactive web apps using R and Python.

In case you missed it in all the posit::conf 2025 content, checking out part 3 of Veerle van Leemput's series on building your first AI powered app is worth checking out! shiny.posit.co/blog/posts/s...

Shiny - Talking to LLMs: From Prompt to Response Shiny is a package that makes it easy to create interactive web apps using R and Python.

Part two of Veerle van Leemput's exploration of LLMs is out. Her exposition is so clear and easy to follow even if you are not likely to use Shiny its worth checking out! shiny.posit.co/blog/posts/s...

Obviously this does not replace manual EDA... but it lowers the bar to taking a quick look so much engaging with data you would not otherwise becomes a real option.

Collaborate > automate. Databot logo alongside screenshot of Positron.

Introducing Databot: an AI assistant for exploratory data analysis in #Python and #RStats!

A research preview in Positron, Databot is a tireless pair programmer to help you explore data.

Learn more about this tool and our philosophy behind it:

🤖 posit.co/blog/introdu...
⚠️ posit.co/blog/databot...

Posit Posit's latest glimpse newsletter highlights new product releases, including the Positron IDE and several new LLM-powered tools.

Finally got Glimpse out focused on what the Posit open source team has been up to. Turns out its quite a bit! posit.co/blog/posit-g...
I continue to be happy with the approach we (Posit) are taking with LLMs and AI. Too impactful to ignore but focus on useful and safe as much as possible.

You might be in a certain kind of family if... as leftovers are being put away in the fridge there is an involved discussion on if the fridge is LIFO or FIFO and if that is family member specific. Then the undeniable trump card is played and its really MDFO (Most Delicious) for everyone.

This is fantastic! Really looking forward to using this … like tomorrow! Let’s gooooo!

Yahs!!!!!!!

Shiny - What LLMs Actually Do (and What They Don’t) Shiny is a package that makes it easy to create interactive web apps using R and Python.

Announcing a new blog series on LLMs from @veerle.hypebright.nl!

Part 1: "What LLMs Actually Do (and What They Don't)" breaks down complex concepts like transformers, hallucinations, and RAG in simple terms.

Check it out here: shiny.posit.co/blog/posts/s...

#RStats #LLM #PythonProgramming

Posit This blog post evaluates various large language models (LLMs) for their Python code generation capabilities, focusing on Pandas-related problems from the DS-1000 dataset.

Testing work or apps that leverage LLMs is disturbingly difficult. At Posit we have been exploring how to do this as we explore RESPONSIBLE usage of LLMs in data science. Sara and Simon share some findings about which models play nice with Pandas. 🐼

posit.co/blog/python-...

RECIPE PLEASE! Brown butter is a cheat code and adding milk solids is ... a double?

RAGs seem to be pretty easy to do _poorly_, but sensible defaults and a smooth path to genuine value is much harder.

Awesome to hear Ragnar performed well.
How much of a lift was it to get those good results with tidymodels docs is the real question.

Ever wonder about the mind behind Pandas & Apache Arrow? 🤔 Ep. 2 of #TheTestSet (Part 1!) unpacks @wesmckinney.com's journey, including his speedrunning past! What makes good tools good?

🎧 Listen at thetestset.co, on Spotify, or Apple Podcasts

#Python #DataEngineering #OpenSource #PodcastLaunch

Let’s be clear. LLMs do not hallucinate.

LLMs are not trying to be correct, they are trying to be plausible.

Posit Drew Coughlin demonstrates how the querychat package can create an LLM-powered dashboard that allows natural language queries of trail race data.

Love seeing the communities work on Posit's blog. If you have an awesome Quarto or Shiny blogpost you think you are missing ping me and lets see if we can work it out! posit.co/blog/race-st...

The image shows a section of documentation for the Iconify Quarto Extension version 3.0.0. The documentation includes a code block with YAML configuration settings for the extension, specifying the title, subtitle, format, and iconify settings such as set, size, and flip. Below the code block, there is a table with two columns: "Shortcode" and "Icon." The table lists four shortcodes for different emojis and their corresponding icons. The shortcodes are '{{< iconify exploding-head >}}', '{{< iconify anguished-face >}}', '{{< iconify astonished-face >}}', and '{{< iconify face-with-spiral-eyes >}}'. The icons are an exploding head emoji, an anguished face emoji, an astonished face emoji, and a face with spiral eyes emoji, respectively. All icons are upside down.

Quarto Iconify 3.0.0 is there!

Set your own global default values via the YAML frontmatter because sometimes you want things to get upside down all at once 🙃

github.com/mcanouil/qua...

#Quarto #Iconify #HTML

So happy to work at a company that shows up for open source! Lets go!!!

Posit Posit's internal AI/LLM hackathons are providing valuable firsthand experience for employees exploring the technology. After several months, here are some key lessons learned.

Love it, or hate it LLMs are going to have a profound impact on software development and data science. At Posit I have been helping the team grapple with the tech. I wrote up our process and experiences with it. Would love to hear about other orgs have engaged with them. posit.co/blog/llm-hac...

Posit Read about new LLM tools for R and Python, improved interoperability, community events, learning resources, and more.

Posit has been up to lots of exciting stuff. I wrote a short overview of it as a Glimpse Newsletter. Lots of LLM/AI (its important whether you love it or hate it)... Beyond product announcements, there are also some great learning resources. Let me know what you think. posit.co/blog/posit-g...

Simon has been killing it, building packages and exploring the use of AI/LLMs in R. It seems like deciding on which base model to use should be straightforward forward, but even that takes thoughtfulness, a systematic approach and is a (fast) moving target!

A ggplot2 bar plot showing the counts of correct, partially correct, and incorrect answers from various LLMs on R coding problems. Claude Sonnet 3.7, shown in orange, answer questions correctly much more often than the GPT series models, shown in shades of blue. GPT-4o answers questions correctly less often than any of the GPT 4.1 series models.

New on my blog: OpenAI dropped a trio of models yesterday that they claim result from "focusing closely on real-world developer needs." I was curious whether this might mean stronger #rstats coding performance.

www.simonpcouch.com/blog/2025-04...

ellmer hex, consisting of a colorful cartoon elephant

We are delighted to announce the release of ellmer, an R package for accessing the power of LLMs in R!

ellmer supports a variety of model providers. Chat, tool call, stream API calls, extract structured data, and more!

Learn about it here: posit.co/blog/announc...

#RStats #AI #LLM

Meet gander, a chat interface for R data scientists🪿

gander automatically incorporates context from your #rstats environment and surrounding code and can be used with any model supported by ellmer in either RStudio or Positron.

Read more: simonpcouch.github.io/gander/

Unified branding across Posit tools with brand.yml - Posit Introducing brand.yml: define your organization’s branding in a single YAML file and apply that branding across reports, dashboards, and apps.

Messing with CSS and doing it project by project, and format by format, across bigger organizations has been a nightmare. I have been a huge flag carrier at Posit for brand.yml to work help solve it. Sara walks us through the first steps on the road.

posit.co/blog/unified...