The depth-first search pattern

The depth-first search pattern: Exploring trees and graphs
15 minutes by Franco Fernando

Depth-first search is a recursion pattern for exploring trees, graphs, and matrices. It works by going as deep as possible along one path before backtracking and trying another. To avoid infinite loops in graphs, you track visited nodes and skip them on future visits. Depth-first search fits problems where you need to find paths, check reachability, or explore all possibilities through a structure.

Stop re-explaining your codebase to your AI agent every morning
sponsored by Jolli

Coding agents are stateless, your context isn't. Jolli Memory stores agent memory as Git-backed files in your repo, so it persists across sessions and follows you between Cursor, Claude Code, Codex, and VS Code. Ships with the codebase to every teammate. Free and open source.

Everyone should know SIMD
11 minutes by Mitchell Hashimoto

SIMD lets a CPU process multiple values at once instead of one at a time, turning slow byte-by-byte loops into chunk-based operations that run 4x to 16x faster. Despite its reputation for complexity, common SIMD code follows the same five steps every time: broadcast constants, loop over chunks, operate on all values at once, reduce the result, then handle leftovers with a normal loop. Once you learn that pattern, writing SIMD is nearly as simple as writing a regular loop.

htop explained
43 minutes by Pēteris Ņikiforovs

Linux tools like htop and top expose a lot of system data, and Pēteris walks through what each number and column actually means. Load average is not simply CPU usage — it counts all running and waiting processes, including those stuck on disk or network. Memory columns like VIRT and RES can be misleading, since virtual memory includes mapped but unused space. Pēteris also covers process states, signals, niceness, scheduling, and what each background service on a fresh Linux server actually does.

In-house LLM serving at Netflix
12 minutes by Netflix Technology Blog

Netflix runs its own LLM serving stack on top of vLLM and NVIDIA Triton, integrated into existing production infrastructure rather than a separate system. Key decisions around engine choice, model packaging, API design, and deployment strategies each revealed unexpected trade-offs only under real production load. A notable example is constrained decoding, where per-request CPU processing caused latency to grow linearly with batch size until a rewrite using vLLM V1's batch-level API and C++ solved the bottleneck.

I inspected my take-home interview project. It was a whole operation.
9 minutes by Appaji C

A fake recruiter on LinkedIn sent Appaji a Python developer job offer with a take-home coding test hidden inside a zip file. The project looked clean at first, but running a simple directory listing revealed pre-loaded Git hooks designed to silently download and execute malicious scripts the moment any Git command ran. The scripts then installed Node.js in the background and ran obfuscated code targeting crypto wallets, with unique tracking IDs assigned to each victim.

And the most popular article from the last issue was:

• How Claude code works, from tokens to agents