Towards Data Science

Follow along the latest deep dive from Nikhil D. to learn about gradient descent and stochastic gradient descent — from their mathematical underpinnings to their real-world use cases.

MCP servers promised structure, but flexibility is proving more scalable. Tomaz Bratanic breaks down why terminal-based workflows are becoming the preferred option for agents.

"RAG systems do not fail only on quality. They can also become inefficient in terms of cost, often in ways that are not immediately visible." Emmimal P Alexander presents a cost-guardrail layer you can build into your RAG system.

The role of a data scientist is changing faster than most people realize. Haden Pelletier highlights three Claude skills that are becoming essential in 2026.

Ready to explore the nitty-gritty details of enterprise document intelligence? Don't miss Angela Shi's deep dive on building a baseline RAG pipeline.

Proof assistants are changing how programmers interact with mathematics. Ronen Lahat explores Lean through the perspective of software engineering and machine learning.

Prompt engineering alone breaks down quickly in production systems. Emmimal P Alexander explains how a control layer can enforce reliable structured outputs without changing prompts.

"This is evaluation of evaluation (eval-of-eval). Instead of only asking whether the AV model is correct, we ask whether the evaluator itself is stable, calibrated, bias-resistant, and useful for downstream decisions." Huma Shah walks us through her innovative work on a framework to stress-test and denoise LLM evaluations workflows.

Guangrui Xie shares an introduction to their open-source ORPilot project, and explains why there's still a gap between operations research approaches and LLMs' attempts to leverage them for real-world use cases.

Based on his recent work with local LLM agents, Hussen Mohammed Ibrahim shares practical insights for making them — and the infrastructure powering them — more useful and effective.