From 0e874a34893c728fc9e142e714b552d84b1b85a1 Mon Sep 17 00:00:00 2001
From: Richie Cahill <Richie@tmmworkshop.com>
Date: Fri, 12 Jun 2026 13:34:59 -0400
Subject: [PATCH] improved queary for vector search

---
 python/ebook_search/search.py | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/python/ebook_search/search.py b/python/ebook_search/search.py
index e25c17d..5c179db 100644
--- a/python/ebook_search/search.py
+++ b/python/ebook_search/search.py
@@ -93,13 +93,14 @@ def search_ebooks(
 
     logger.info("ebook_search_start query_length=%s rerank=%s", len(query), rerank)
     timings: list[RuntimeStep] = []
-    retrieval_query, timing = timed_result("Query preparation", retrieval_query_from_text, query)
+    bm25_query, timing = timed_result("BM25 query preparation", retrieval_query_from_text, query)
     timings.append(timing)
     retrieval, timing = timed_result(
         "Hybrid retrieval",
         parallel_retrieval,
         engine,
-        retrieval_query,
+        query,
+        bm25_query,
         config,
     )
     timings.extend(retrieval.timings)
@@ -130,7 +131,12 @@ def search_ebooks(
     return response
 
 
-def parallel_retrieval(engine: Engine, query: str, config: EbookSearchConfig) -> RetrievalResponse:
+def parallel_retrieval(
+    engine: Engine,
+    vector_query: str,
+    bm25_query: str,
+    config: EbookSearchConfig,
+) -> RetrievalResponse:
     """Run vector and BM25 candidate retrieval concurrently with separate database sessions."""
     with ThreadPoolExecutor(max_workers=2, thread_name_prefix="ebook-search") as executor:
         vector_future = executor.submit(
@@ -138,14 +144,14 @@ def parallel_retrieval(engine: Engine, query: str, config: EbookSearchConfig) ->
             "Embedding + vector search",
             vector_candidates,
             engine,
-            query,
+            vector_query,
             config,
         )
         bm25_future = executor.submit(
             timed_result,
             "BM25 search",
             bm25_candidates,
-            query,
+            bm25_query,
             config,
         )
         vector_results, vector_timing = vector_future.result()
@@ -196,7 +202,7 @@ def apply_rerank(
 
 
 def vector_candidates(engine: Engine, query: str, config: EbookSearchConfig) -> list[SearchResult]:
-    """Return pgvector cosine candidates for a normalized query."""
+    """Return pgvector cosine candidates for a natural-language query."""
     with Session(engine) as session:
         model = session.scalar(select(EbookEmbeddingModel).where(EbookEmbeddingModel.name == config.embedding_model))
         if model is None: