Building a Real-Time Context Engine for Our Code LLM: No More 10-Minute Delays