End project: a Local LLM Test Generator (React + Express) where you paste a Jira story like VWO-48 and get strict Jira-format test cases back, from llama3.1 running on your own laptop via Ollama, or a cloud model, switched with one dropdown. Press Play to see why raw models drift and how guardrails fix it.