| | Ready to hone your skills? Fidelity can provide the platforms and dedicated time needed for mastering new technologies. Enjoy the curiosity of a startup with the stability of a private firm that cares about helping you grow. Discover your tech future at Fidelity.
*Message from this week's sponsor, Fidelity Careers |
| | |
| | 🔗 Note: I’m pretty sure the links-not-displaying issue is completely resolved. But if you don’t see links in this email, please reply and let me know. For now, the newsletter is also available to read online: dataelixir.com/archive/issue-574.html |
| | |
| | This post starts as a project about tagging old blog posts, but the useful part is the workflow. BERTopic handles the clustering, LLMs help refine the taxonomy, Claude Code applies tags across the repo, and human judgment keeps the whole thing coherent. It's a good example of AI helping to finish a small, well-scoped data project.
Vicki Boykis |
| | |
| | Cross-country regressions can make big social patterns look cleaner than they are. In this post, Julia Rohrer explains why it’s so easy to overinterpret country-level moderation effects, from small sample sizes and confounding to measurement issues, then offers better ways to make the work more transparent and defensible.
The 100% CI | Julia Rohrer |
| | |
| | Figuring out the world’s longest river sounds like a simple measurement problem until you really look at it. This post uses the Amazon-vs-Nile debate to show how definitions, source data, and measurement choices can turn a basic fact into a surprisingly messy data question.
Jonn Elledge |
| | |
| | YAML gets blamed for a lot of strange config-file behavior, often for good reason. But this post makes the case that the real problem is usually old YAML 1.1 tooling, not modern YAML itself. This is a useful tour through parser behavior, TOML tradeoffs, and why config formats still matter.
Posit Blog | Rich Iannone, Tomasz Kalinowski |
| | |
| | AI is speeding up analytics work, but the fundamentals still decide whether anyone trusts the output. dbt’s 2026 State of Analytics Engineering Report looks at AI-assisted coding, governance gaps, infrastructure costs, and the growing pressure to deliver reliable insights faster. Learn more.
// sponsored |
| | |
| | | | AI can make a dashboard look polished in minutes, but polish is not the same as analytical judgment. This post tests three Claude Design prompts and shows how AI-generated dashboards can still fail in surprisingly familiar ways.
The Datavist | Darragh Murray |
| | |
| | | Most Python time-series tools either detect change points, flag anomalies, or classify market-style trend segments. PyTrendy is trying to package the analyst-facing version of the problem: identify up, down, flat, noisy, abrupt, and gradual trend segments in a way that is easy to inspect and explain.
GitHub | Russell Sammut-Bonnici |
| | |
| | A webpage is already structured data, just not usually in a form you can query. SiteRows treats page elements like tables, so you can use SQL to pull links, headings, images, metadata, forms, and lists. It’s early, but “query a website with SQL” is an interesting idea.
SiteRows |
| | |
| | | GPUs are the machinery behind modern ML, but most of us only see them as expensive boxes that make things faster. BrrrViz opens that box with interactive visual explanations of GPU programming concepts, from chip architecture and execution hierarchy to memory, synchronization, and optimization.
brrrviz |
| | |
| | | This week's issue sponsored by: | | |
|
