Conspiracy Generator
Name a household object. Learn what they are hiding.
What this is: a small, silly web app that writes an absurd, obviously fake conspiracy theory about any household object you type in.
How it works: a Python (FastAPI) backend generates the story with one of two engines. Python templates are instant. Local LLM runs a small Llama 3.2 model on the owner's ancient home server through Ollama, which is slow on CPU and falls back to templates if it times out or refuses.
Why it exists: it is a sandbox for practicing application monitoring with New Relic. The toggles below create problems on purpose so they show up in the dashboards: Slow mode adds latency, Error mode fails about one request in five and Heavy load spikes the CPU. LLM calls are also recorded so model speed and fallbacks can be charted.