Polishing an Electron app's UI with an AI design skill on the front and a code knowledge graph on the back: design vocabulary plus a real source of truth.

An AI agent investigating a slow server found a Monero miner consuming 50% CPU inside a Docker container. Here's the full incident — how it got in, what it was doing, and everything wrong that allowed it.

Spent half a session tuning VLM localize prompts. The real bug was deep in my RFB client: framebuffer captures were 1–N frames behind reality.

Merging turbo3 KV cache and MTP speculative decoding into one llama.cpp binary: the build crashes, the CUDA dispatch bug, and 252K context at 85% draft acceptance.

Practical comparison between Qwen 3.6 Dense and MOE on an RTX 3090, focused on real throughput by scenario and the practical impact of Multi-Token Prediction in local inference flow.

Testing Unsloth's Qwen 3.6 27B MTP GGUF on an RTX 3090 with llama.cpp's MTP branch: native speculative decoding, no draft model, real speedup.

How a single flag in llama.cpp turns a 35B Mixture-of-Experts model from OOM to 23 tok/s on a 12GB GPU.

Benchmarking Multi-Token Prediction (MTP) on Qwen 3.6 27B via llama.cpp on an RTX 3090 — 1.5× speedup in agentic tool-call chains.

A web dashboard for 96 PR reviews — built with Python stdlib, Chart.js, and zero external server dependencies.

After building an LLM-powered issue manager, the natural next step was PRs — syncing, diffing, reviewing, and posting feedback with the same local-first pattern.