| | At Fidelity, you can work on cutting-edge projects that shape the future of the tech industry —while still enjoying your time outside the office. A culture that supports your well-being empowers you to do your best work, which is why Fidelity offers on-site wellness centers, personal assistance through a concierge service, and almost 5 weeks of paid time off to all associates. Learn more about how we approach technology differently at Fidelity. Terms and conditions apply to all benefit programs.
*Message from this week's sponsor, Fidelity Careers |
| | |
| | 🔗 Note: if you don’t see links in this email, it’s also available on the web: dataelixir.com/archive/issue_578.html |
| | |
| | Data teams often measure success by what they ship: pipelines, dashboards, reports, tickets closed. This post argues that the better measure is whether any of it changes decisions. The useful framing is Data -> Perspective -> Action, with practical habits for turning analysis into recommendations, narrative, and organizational influence.
Practical Data Community | Goutham Budati |
| | |
| | Coding agents are less mysterious once you look at the tools they’re allowed to use. In this post, Hadley Wickham breaks them down into a small set of functions for reading files, editing code, and running commands, showing how tools like Claude Code and Codex turn an LLM into something that can actually work inside a codebase.
Tidy design principles | Hadley Wickham |
| | |
| | MNIST is the classic “hello world” dataset for computer vision, but it’s often too simplistic to expose the interesting differences between neural networks. In this short video, Vincent Warmerdam explains MNIST-1D, a tiny benchmark that turns digit classification into noisy one-dimensional signals. It’s lightweight, surprisingly hard, and it’s useful for demonstrating ideas like inductive bias and double descent. This is a great explainer.
YouTube - marimo and alphaXiv channel | Vincent D. Warmerdam - 9 mins |
| | |
| | Modeling time in databases gets messy fast. Postgres 19 adds native temporal table support, including overlap-free primary keys, updates that apply to only part of a time range, and foreign keys that understand validity periods. This post shows why that matters with concrete examples.
pgEdge | Shaun Thomas |
| | |
| | SQL Concepts Lab is an interactive SQL trainer that runs entirely in the browser using DuckDB-WASM. It includes ten lessons, runnable examples, checked exercises, and a small sample dataset designed to make joins, NULLs, windows, EXPLAIN, and dialect differences concrete.
Örjan Lundberg |
| | |
| | A lot of data journalism stops at the article. Recovered Factory is pushing further by publishing the workflow, building reusable tools, showing the methodology, and making the data accessible through AI tools like MCP servers. Their newsletter is a good fit for Data Elixir readers who like seeing how serious data projects actually come together.
// sponsored |
| | |
| | | | Nice collection of 100 real data science projects that helped candidates get callbacks, including the problem, method, metric, and company signal. It’s useful both as a source of project ideas and as a guide to making portfolio work feel concrete and resume-ready.
BuildML |
| | |
| | | supertree makes decision-tree visualization interactive inside Jupyter, JupyterLab, and Colab. It supports scikit-learn, XGBoost, LightGBM, and ONNX models, with zooming, panning, collapsible nodes, and sample-path tracing. Handy for inspecting trees, explaining model behavior, or making tree-based models easier to discuss with others.
GitHub | mljar |
| | |
| | xan is a fast CSV-wrangling tool, but this cookbook shows its overlooked superpower: quick visual exploration directly in the terminal. It can render tables, reports, histograms, scatter plots, heatmaps, and sparklines from the command line, making it handy for fast inspection before opening a notebook. For readers who like lightweight, scriptable workflows, this is surprisingly useful.
médialab Sciences Po |
| | |
| | | I wrote this up after building a local, file-based research assistant for my own investing workflow. It watches my portfolio, reads market news and filings, keeps markdown memory, and sends me a daily briefing. The investing angle is personal, but the more interesting part is the agent design: small specialized agents, local files, editable memory, and a synthesis step that turns ongoing information flow into something useful.
Lon Riesberg |
| | |
| | | This week's issue sponsored by: | | |
|
