The port-operations copilot for Meridian's terminal planners retrieves the most recent berth-allocation circulars and grounds its answer in them, but planners currently see raw retrieved passages glued in front of the model's answer in the chat window. Which chain component is missing?
Select an answer to reveal the explanation.
Short Explanation
You’re seeing retrieved passages appear raw in the chat, so the missing piece is an output-formatting step that cleans up the model’s response before it reaches the planner. Think of it as an editor turning a draft into a clear final copy.
Full Explanation
Raw retrieved passages appearing in the planner’s chat window point to a presentation problem after generation, not a failure to find or retrieve information. An output-parsing/formatting step between the LLM and the UI can process the model’s response, remove or organize unwanted material, and pass a clean answer to the planner. A vector index refresh job addresses stale or changed source documents; it does not change how an already generated response is displayed. A larger embedding model may affect which passages retrieval finds, but it cannot format the response after retrieval has succeeded. A reranker can prioritize the most relevant berth circulars before they reach the model, which may improve relevance, but it does not clean up raw passages in the UI. The exam caveat is to distinguish retrieval quality from answer presentation: when the retrieved material is current and grounds the answer, changing retrieval components does not address a display-stage symptom. As a practical check, inspect the chain after the LLM and before the UI to see whether a parsing or formatting step is missing or bypassed.