Large documenet sets often run beyond the limited length of LLM context window. Additionally given the diverse topical landscape in a large corpus of documents, injecting them all into a single context might not produce the desired results. We have developed an approach that uses a topical ontology to first cluster documents by topic around central concepts of the corpus prior to LLM based summary generation. We show significant improvemnts in terms of topical precision and recall over ChatGPT Deep Research in summarizing the daily news in politics for Politics March 14th 2025.