Writing & Ideas
Thoughts on AI engineering, LLM systems, and building software that scales.
Why Your FastAPI Endpoint Blocks the Event Loop
“Just make it async def” is the advice every FastAPI developer hears first, and it’s the reason so many production incidents start …
ai-memory: Persistent, Cross-Tool Memory for AI Agents
You quit Claude Code mid-task after four hours of back-and-forth — architecture decisions made, three approaches tried and abandoned, one open …
Inside X's Bot Detection: botmaker, scarecrow, bdsm
On August 13, 2026, X pushed an update to its open-source x-algorithm repository that most coverage reduced to one headline: the exact ranking weights …
Needle: Function-Calling AI on Tiny Devices
A post about a 26-million-parameter open-source model has been circulating on LinkedIn: Needle, from Cactus Compute, distilled from Gemini for …
Understanding QUIC: How HTTP/3's Transport Works
Most backend developers can still sketch TCP’s three-way handshake from memory, but ask about QUIC and the conversation stalls out at …
Open Knowledge Format (OKF): A Self-Updating Graph
Point a coding agent at a real repository and watch what actually happens before it writes a single line: it reads files, greps for callers, opens a …
Twilio Telephony Terms: SIP Trunking, DIDs, A2P
The first time you open Twilio’s console to wire up voice or SMS for a real product, the terminology hits you all at once — SIP trunk, DID, …
UUIDs Explained: Versions, Collisions, Performance
Ask most backend developers what a UUID is and you’ll get the same answer: “a random string that’s basically unique, so I …
Ruff in Python: Replace Your Lint and Format Stack
Most Python projects don’t have one code-quality tool — they have five. Flake8 for style, Black for formatting, isort for imports, pyupgrade for …
MCP Servers Explained: Deploying One to AWS EC2
Most MCP content is theoretical — diagrams of a client talking to a server, an analogy to USB-C, a five-line “hello world” that runs on …
RAG in Production: Why It Works in Demos, Fails Live
The demo always works. You embed a few dozen documents, ask a question, and the model answers it correctly, instantly, with a clean citation. You show …
File Upload Security: A Complete Guide
Every application that lets users upload something — a profile picture, a resume, an invoice, an attachment — opens a door into your server. Most of …