Data Science Chronicles
953 articles on models, math and shipping ML to production. Press/ to search, or explore by topic and year.
We’ve accepted that Python is slow for decades. But running sandboxed Python inside WebAssembly just flipped the script on CPython—and the benchmarks will shock you.
Meet Kolibri, the sovereign German LLM that is rewriting the rules of European AI by trading raw parameter bloat for uncompromising explainability, data privacy, and trust.
We’ve been brainwashed into thinking every business problem requires a deep neural network. It's time to talk about why deterministic decision models are quietly crushing ML classification in production.
Discover how pretrained deep learning intent signals are completely rewriting the rules of mobile programmatic advertising, leaving legacy heuristics in the dust.
As neural networks begin cracking centuries-old conjectures and generating novel proofs, the pure math community faces an existential reckoning. Are we witnessing the dawn of automated abstraction?
For years, the gospel of artificial intelligence was simple: just add more compute and more data. But behind closed doors in Silicon Valley, the industry's heaviest hitters are sounding the alarm. The scaling wall has arrived.
Autonomous AI agents are rewriting enterprise workflows, but managing their OAuth access tokens is a ticking time bomb. Here is how Amazon Bedrock AgentCore solves user consent at scale.
Traditional machine learning pipelines are officially dead. Discover how Instacart uses autonomous agentic loops to self-heal models, slash compute costs, and reinvent grocery delivery at scale.
Your polyglot data science pipeline is a ticking time bomb of broken dependencies. Here is how modern engineers are fixing it once and for all.
We’ve accepted that Python is slow. We’ve accepted that running untrusted code is a security nightmare. Edge Python in WebAssembly just shattered both of those assumptions.
Another top safety leader has walked out of OpenAI, slamming its corporate culture as 'broken.' Here is what the high-stakes friction between commercial velocity and AGI alignment means for the future of AI architecture.
If your multi-language data pipeline breaks every time a database updates, you're doing polyglot data science wrong. Here is the modern architectural blueprint to fix it forever.
Recommended
The three resources I recommend most to anyone getting into data science.
Aurélien Géron's practical tour from linear models to deep learning.
Why I recommend itThe book I recommend most to people starting out — every chapter ends in runnable code.
View on Amazonon amazon.comChip Huyen on taking models from notebook to production.
Why I recommend itThe best single read on MLOps, data drift and monitoring that I know of.
View on Amazonon amazon.comStanford × DeepLearning.AI's beginner-friendly ML course on Coursera.
Why I recommend itStill the gentlest, clearest on-ramp into ML I've seen.
Start learningon coursera.orgSome links are affiliate links: if you buy through them I may earn a small commission, at no extra cost to you. I only recommend things I've used or would recommend to my own team. Details