Add rerank module: bi-encoder retrieval plus an ONNX Runtime cross-encoder reranker behind a Spring AI DocumentPostProcessor; hit@k and MRR by pool size and question kind, latency by candidates, batch size, threads and int8, and five reproduced traps. Local MiniLM models, no database or API key.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Cfo2K4u4rF18QsPXPFnytf
2026-10-24 18:00:05 +00:00
parent 2ccf1d8d19
commit a429aa40f6
34 changed files with 1617 additions and 0 deletions
@@ -0,0 +1,11 @@
+# Bi-encoder alone against bi-encoder top 50 + cross-encoder, by kind of question
+
+kind                   ranking    hit@1    hit@3    hit@5    MRR@10
+exact error code       bi-encoder  2/12   5/12   8/12  0.374
+                       reranked    8/12  12/12  12/12  0.792
+exact SKU or setting   bi-encoder 12/12  12/12  12/12  1.000
+                       reranked   12/12  12/12  12/12  1.000
+paraphrase             bi-encoder 15/24  22/24  23/24  0.767
+                       reranked   17/24  20/24  20/24  0.778
+code plus words        bi-encoder 12/12  12/12  12/12  1.000
+                       reranked   12/12  12/12  12/12  1.000