Summarizer-to-Classifier Chain With GPT-4o-mini

Decomposing a task into multiple focused LLM calls—specifically a summarizer call followed by a classifier call—is usually more reliable and cheaper than one giant prompt trying to do both.

Call 1 (Summarizer): Summarize the key clinical findings from this 30-page trial report in 150 words. Call 2 (Classifier): Based on the 150-word summary above, classify the trial outcome as [PRIMARY_ENDPOINT_MET, PARTIAL_EFFICACY, FAILED_ENDPOINT].