Alex Kraieski
LightNews LightNews
banner
alexkraieski.bsky.social
Alex Kraieski
@alexkraieski.bsky.social
250 followers 370 following 490 posts
Laravel developer, data scientist, and accessibility advocate
🐘 TALL stack | PostGIS
📈 R + Tidyverse | Shiny
NH, USA | Clemson University alum (Poli Sci)
DM for freelance/consulting availability or to chat about whatever!
https://kraieski.dev
Posts Replies Media Videos
alexkraieski.bsky.social
I'm curious, does anyone have any interpretations of the results from Staten Island specifically that they're willing to share?

I'm thinking of making some visualizations of my own about this, but my knowledge about NYC political culture has limits too
NYC mayoral election results for Staten Island. Cuomo 55.4%, Mamdani 22.7%, Silwa 21.3%
November 6, 2025 at 1:17 AM
alexkraieski.bsky.social
"Hey ChatGPT, please help me tear a pec"

This ChatGPT ad on YT is clearly promoting it as a fitness/health tool. Yuck.
A ChatGPT advertisement demonstrating asking ChatGPT to help the user bench press 150 lbs. by the end of the year.
October 7, 2025 at 9:59 PM
alexkraieski.bsky.social
The evidence is pretty weak IMO that crime in DC is even trending upward in the long term or short term. Many forms are down from a peak in 2023.

Any nominal growth in homicides is not statistically significant.

Let's be honest, Trump's DC takeover is about imposing racism, not reducing crime.
Line chart of homicides in Washington, DC from 2005 to 2024. Although homicides have been rising since 2012, it is unclear that a long-term uptrend exists, and homicides fell off drastically from a high in 2023 Line chart of homicide rates in Washington, DC from 2005 to 2022. The trend appears relatively flat over time, albeit with some volatility. Notably, progress in lowering the homicide rate was undone during the first Trump Administration.
August 12, 2025 at 5:51 PM
alexkraieski.bsky.social
I challenged Claude to make a ggplot2 theme based on 1993's Doom with my ggthemeAI #rstats package. I think it did a decent job!

I recently added some goodies for starting {ellmer} chats: ggplot2 expert system prompts w/ embedded documentation and a tool to let the LLM see what fonts are available
June 27, 2025 at 1:32 AM
alexkraieski.bsky.social
Wish there was an #F1 race this weekend. But I guess that means I can post a little more about last weekend's Canadian GP. Here's a lap time comparison I made. Russell was particularly strong with his late-stint pace.
Line chart comparing lap times of Russell (cyan) and Verstappen (orange) during the 2025 Canadian Grand Prix. Lap times are mostly similar but show sharp spikes for pit stops around laps 12 and 40ish and a safety car at the end.After the first stint, the lap times tend to trend down throughout the race for both.
June 21, 2025 at 1:45 AM
alexkraieski.bsky.social
I've been having fun with using {ellmer} to generate #ggplot2 theme functions

claude-sonnet-4 and gpt-4.1 both have their strengths and weakness for this, but I thought this 80s theme from gpt-4.1 looked pretty cool!

working on an R package for my 'make_ai_theme()' function

#rstats #ai
a histogram with bright colors: Blue, pink, and yellow r code: library(ellmer) make_ai_theme <- function(chat, theme_prompt){ prompt <- interpolate_file('custom-theme-prompt.md', theme_request = theme_prompt) # intentional side effect: you can chat w/ the LLM about the theme later theme_text <- chat$chat(prompt) expr <- parse(text = theme_text) theme_function <- eval(expr) theme_is_function <- is.function(theme_function) if (!theme_is_function){ stop('The response from the LLM wasn\'t an R function. Maybe you should yell at it. Or try a different LLM provider/model') } theme_function }
June 14, 2025 at 2:12 AM
alexkraieski.bsky.social
I've been working on an interactive chart gallery for my website, and accessibility has been a big focus. Can be a challenge with JS data viz libraries

For screen readers, I used the ARIA 'img' role and set 'aria-label' to a concise text description. seems to work well with VoiceOver
#accessibility
screenshot showing part of a website UI with a chart and a sidebar where users can browse through a list of charts to view. On the screen is the "window" where the VoiceOver screenreader shows what it is reading. It demonstrates that the plot is accessible to users who use a screen reader.
June 13, 2025 at 3:35 AM
alexkraieski.bsky.social
Google's Pagespeed Insights wants me to "Avoid enormous network payloads" to make a webpage load faster. The enormous payloads are #rstats binaries from #webR. I will not comply!

This is for a page on my own site and it scores 40 for speed on mobile. NY Post homepage is even slower, I checked!
a partial screenshot from Google's pagespeed insights with the suggestion "Avoid enormous network payloads." over 6,000 KiB of this is from WebAssembly binaries for R, the statistical programming language.
June 12, 2025 at 6:47 AM
alexkraieski.bsky.social
Here's a functional programming trick for #rstats that I wish I started using sooner:

if you need a #ggplot2 scale to be reusable across multiple plots and dynamically configurable without relying on global state, consider using a function factory (a function that returns a function) to build it
screenshot of a code editor showing the following R code: library(quantmod) library(ggplot2) library(lubridate) startYear <- 2015 startDate <- paste0(startYear, '-01-01') getSymbols(c('spy', 'btc-usd'), from= startDate) # function factory that creates a scale function that only shows valid years. # try to keep code that could change in here! make_valid_year_scale_function <- function(start_year){ function(){ list( scale_x_continuous(breaks = seq(start_year, Sys.Date() |> year(), 1)), theme(panel.grid.minor.x = element_blank()) # use function after other theme funcs ) } } # this makes it so I can add scale_x_valid_years() to any plot scale_x_valid_years <- make_valid_year_scale_function(startYear)
May 29, 2025 at 11:36 PM
alexkraieski.bsky.social
Woah this seems like a big deal!

#RStats people, do you like using GitHub Copilot in RStudio?

I am on the fence... it is frequently a distraction imo, but this update should improve quality
Lastly, RStudio now uses the RStudio project folder as the GitHub Copilot workspace root. Copilot suggestions are now more relevant, utilizing your project folder context.
May 8, 2025 at 9:08 PM
alexkraieski.bsky.social
And finally, just to tie in where the CMS part comes in, I have a secure control panel for adding/updating plots with the form shown here

Then, I can query this collection in the PHP backend to display on my site.

And generally, I think webR exposes R more to what various web frameworks do well
A screenshot showing a form UI in a CMS for entering R code for data and a plot that will be viewed on the website's fontend
May 7, 2025 at 8:43 AM
alexkraieski.bsky.social
Just finished building a live webR #ggplot2 gallery for my website! It actually runs #rstats on the frontend to generate example charts (w/ help from plotly.js)

Backend is Statamic (CMS built with Laravel), and I connect it to the frontend with Livewire. Shiny server not required
May 7, 2025 at 2:59 AM
alexkraieski.bsky.social
Yeah, I'll let you know if/when I get back to something here, thank you! I did confirm with a quick test that I am able to pull "Truths" via api, but setting a 'limit' greater than 20 doesn't have any effect.

Hopefully there's a way around that 20 record restriction!
A snippet R code to load Donald Trump's "Truths" library(httr2) req <- request('https://truthsocial.com/api/v1/accounts/107780257626128497/statuses?exclude_replies=true&with_muted=true') # change user agent req <- req_user_agent(req, "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.3") resp <- req_perform(req) data <- jsonlite::fromJSON(resp_body_string(resp), flatten = TRUE)
April 16, 2025 at 6:59 AM
alexkraieski.bsky.social
Visualized: Luka Doncic appears to be shooting more free throw attempts per game since being traded to the Lakers

#NBASky #dataviz
Boxplot comparing Luka Dončić’s free throws attempted per game in 2025 with two teams: the Dallas Mavericks and the Los Angeles Lakers. The Mavericks boxplot is in dark blue, showing a median around 6, with a wider interquartile range and two high outliers near 15 and 16 free throws. The Lakers boxplot is in orange, with a higher median around 9 or 10 and a more compressed distribution overall, extending from about 3 to 13 attempts. This suggests Dončić attempted more free throws per game, on average, when playing for the Lakers compared to the Mavericks.
April 14, 2025 at 6:25 PM
alexkraieski.bsky.social
Shout-out to the F1 qualifying broadcast for casually dropping one of the worst data visualizations I've ever seen

#F1 #JapaneseGP #SkyF1
A color-coded map that attempts to show where on track Liam Lawson and Yuki Tsunoda are fast relative to each other. The color scale shows both as blue, making the map hard to read.
April 5, 2025 at 4:25 PM
alexkraieski.bsky.social
Looking at the data for Perez in his first season for Red Bull (2021) tells a similar story. Even though Red Bull is clearly showing more urgency than they have in the past, it also appears that Lawson has less qualifying pace than other drivers that were inadequate for Red Bull in the past
Line chart titled “Sergio Perez Q1 Performance (2021)” with a subtitle: “W/ Liam Lawson 2025 Comparison.” The chart tracks Perez’s Q1 qualifying pace relative to Max Verstappen over 22 races. The y-axis shows Q1 lap time as a percentage of Verstappen’s (from ~99.5% to 101.5%), and the x-axis represents race numbers 1 through 22. A jagged green line shows Perez’s race-by-race pace. His times mostly range between 100% and 101% of Verstappen’s, with fluctuations and occasional dips below 100%. A black dashed line labeled “Lawson average 2025” sits just above the 101% mark, providing a reference point. Most of Perez’s Q1 performances in 2021 are faster than the Lawson average, with several races showing Perez within 0.5% of Verstappen’s time. One notable spike (race 11) puts Perez close to Lawson’s 2025 average, but overall, Perez generally qualifies closer to Verstappen.
March 30, 2025 at 3:27 AM
alexkraieski.bsky.social
In #F1, I was a little surprised when #RedBullRacing decided to drop Liam Lawson after only 2 races.

I see the case for it though. In 2019, Gasly and Albon were rarely worse than Lawson's average when Q1 lap time is plotted as a percentage of Max's lap time

#sportsanalytics #ggplot2
Line chart titled “RBR #2 Driver Q1 Performance (2019)” showing how Red Bull Racing’s second drivers—Pierre Gasly and Alexander Albon—performed in Formula 1 Q1 qualifying sessions compared to Max Verstappen. The y-axis represents Q1 lap time as a percentage of Verstappen’s time (from 99% to 107%), and the x-axis shows race numbers from 1 to 12. Gasly’s data points (blue line with dots) appear in races 1–12, showing a fairly consistent performance between 100% and 102.5% of Verstappen’s times, with small fluctuations. Albon’s data (red line with dots) begins around race 5 and overlaps with Gasly between races 5 and 9. Albon has one large spike around race 4 (over 106%), suggesting a notable qualifying deficit in that event. Otherwise, Albon generally performs slightly closer to Verstappen’s times than Gasly. A black dashed horizontal line is labeled “Lawson average 2025” and is set at about 101.5%, serving as a benchmark for comparison. Most of Albon’s and Gasly’s data points fall below this line, suggesting Lawson (in 2025) was, on average, further from Verstappen’s Q1 pace than either Gasly or Albon in 2019. A legend below the chart indicates red is for Albon and blue is for Gasly.
March 30, 2025 at 3:15 AM
alexkraieski.bsky.social
Here's my WIP jacket! Trying to keep a strong antifacist vibe! Waiting on some patches for 90s death metal bands and the disability pride flag

#metal #battlejacket #ratm
A denim jacket with several band patches sewn on. On one side, there's a Metallica "Ride the Lightning." On the other side, there are patches for Rage Against the Machine, System of a Down, and Slipknot
March 29, 2025 at 2:42 AM
alexkraieski.bsky.social
I think I've finally achieved the creation of the most cursed Chatgpt code snippet yet.

It's a #perl "one-liner" that implements a #brainfuck interpreter. Piped into that is a message that encourages you to visit my humble website (full command in alt text)

Brain damage confirmed
echo '++++++++++[>++++++<-]>++++++++++++++++++++++++++.+++++++++++++++++++.++++++++++.----------.+++++++++++.------------------------------------------------------------------------------------.+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++.+++++++.-----------------.++++++++.----.++++++++++++++.--------.--.-----------------------------------------------------------.++++++++++++++++++++++++++++++++++++++++++++++++++++++.+.+++++++++++++++++.' | perl -e '$code=<>;@t=(0)x1e4;$p=0;for($i=0;$i<length($code);$i++){$_=substr($code,$i,1);$p++if$_ eq">";$p--if$_ eq"<";$t[$p]++if$_ eq"+";$t[$p]--if$_ eq"-";print chr$t[$p]if$_ eq".";$t[$p]=ord(getc)if$_ eq",";if($_ eq"["){$t[$p]||do{$d=1;do{$i++;$d++if substr($code,$i,1)eq"[";$d--if substr($code,$i,1)eq"]"}while$d}}if($_ eq"]"){$t[$p]&&do{$d=1;do{$i--;$d++if substr($code,$i,1)eq"]";$d--if substr($code,$i,1)eq"["}while$d;$i--}}}'
March 20, 2025 at 4:13 AM
alexkraieski.bsky.social
Excited to get into The Anti-Ableist Manifesto by Tiffany Yu!

I think it's critical to have a framework for confronting ableism. Trump and co. are going to further weaponize society's ableism while trying to maintain plausible deniability. We can't let them. #disability #booksky
Cover of The Anti-Ableist Manifesto by Tiffany Yu. The subtitle says "Smashing Stereotypes, Forging change and building a disability-inclusive world
February 25, 2025 at 6:53 AM
alexkraieski.bsky.social
The Celtics have been shooting well lately. hopefully they can continue that tonight against the Cavs! #NBAsky #rstats
Line plot of the Boston Celtics’ 3-point shooting percentages during the 2024-25 season. Each game is represented by a dot, with blue indicating wins and red indicating losses. A black line connects the points, showing game-to-game fluctuations. A red dashed line marks the season average. The data shows considerable variation, with some games above 50% and others below 30%.
February 5, 2025 at 12:20 AM
alexkraieski.bsky.social
I really love this accessibility feature in Bluesky! If you go to accessibility settings, you can make it force you to add alt text to all images before you post. I recommend that you turn this on if you are serious about reaching as many people as possible! #accessibility
A screenshot of the Accessibility settings on the Bluesky web app showing that users can toggle a setting for "Require alt text before posting." The screenshot indicates that the user has turned it on A screenshot of the Bluesky web app depicting the process of creating a new post. In the screenshot, there are images attached to the post, but there are a number of accessibility warnings. There is a warning that reads "One or more images is missing alt text." The option to actually post is disabled until this is resolved. Lower on the screen, there is an explanation that "Alt text describes images for blind and low-vision users, and helps give context to everyone."
January 17, 2025 at 8:11 AM