Act as a retrieval systems specialist. Build an interactive workbench where I paste a long document and a list of 10 test questions. Chunk the document using five strategies: fixed-size, sentence-based, paragraph-based, recursive, and heading-aware. Let me control chunk size and overlap. For each strategy, simulate retrieval by keyword or TF-IDF scoring and report whether the chunk containing the correct answer was retrieved in the top 1, 3, and 5. Show chunks that split an answer across boundaries, the average chunk length, and the redundancy caused by overlap. Visualize results as a heatmap of questions against strategies. Conclude with a recommended configuration for this document type and explain the tradeoffs in plain language.
0 Comments