Brain Dump
187 / 187 entriesDocker Command Reference
Essential Docker commands for building, running, and managing containers — from hello-world to production deployments.
Ollama vs llama.cpp vs vLLM — picking your LLM serving layer
Three different ways to run LLMs locally and when each one makes sense. I tried all three approaches and ended up on llama.cpp for reasons that only became obvious after a lot of pain.
Kubernetes
kubectl commands for pods, services, deployments, and namespaces — the handful worth keeping open in a terminal.
MoE vs Dense — why a 27B model beat a 122B model on my hardware
Mixture of Experts sounds impressive and the parameter count looks better on paper. But I spent a week learning the hard way that bigger numbers don't always mean better results.
Git
Git commands I actually reach for — clone, commit, branch, stash, and how to undo a mistake with reset and revert.
Tokens, quantization, and context windows — what they actually mean
I kept nodding along when people said 'tokens' and 'quantization' without actually knowing what they meant. Here's what they actually are.
Creating SSL Certificates with OpenSSL
Generate a private key, a CSR and a self-signed certificate with OpenSSL — plus the request you send to a CA when you need it properly signed.
Why LLMs eat VRAM and what actually happens during inference
Finally understood what inference actually means, why GPU VRAM is the bottleneck, and why my 64GB of RAM did almost nothing to help when the model was too big
Building a Local AI-Powered Security Research Agent from Scratch
A full honest build log of setting up a completely local, private AI agent stack for authorized security research and bug bounty — llama.cpp, Goose, Qdrant RAG, model routing, Telegram integration — everything that went wrong and how it got fixed.