From incident reports to connected answers

How can an LLM app connect incident reports and explain a failure across services, instead of just finding a matching paragraph?