This is what happens when you are jobless😅

Just figured out Docker's MCP Gateway - Dynamic MCP feature. It's pretty clever - instead of executing code directly, it: 1. Creates a new dynamic tool (code-mode-{name}) 2. Gives you a JS sandbox with access to MCP server tools 3. You execute it via mcp-exec Example:

Ollama’s default model eviction unloads way more VRAM than needed, causing unnecessary reloads and slowing everything down. So I wrote a small knapsack-based eviction function that picks the minimal subset of models to offload to free the required VRAM. Optimal eviction → fewer

Learning Celery task queue and compiling my own docs.

Completed the Web Researcher: v1 project with a simple streamlit app checkout the repo: https://t.co/NVyqiY0pT5

Just built my first multi-agent system. The idea was to build a very simple version of deep researcher and test it with models served locally with ollama. Sadly 6gb vram limits my model choices. lets see how deepseek-r1:4b performs.