Improve search quality: RERANK_LIMIT 10→60, search_text for dense, variants+hyde multi-vector

- RERANK_LIMIT 10→60: rerank more candidates → better NDCG ordering
- Dense query uses search_text if available (semantically richer than text)
- Sparse query always appends keywords on top of base text
- Multi-vector: use variants[:2] + hyde[:2] as extra dense queries
  (previously only hyde[:2])
- Index UVICORN_WORKERS 8→4: matches 4-core constraint, saves ~1GB RAM

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

index/main.py

@@ -11,7 +11,7 @@ from pydantic import BaseModel
 
 HOST = os.getenv("HOST", "0.0.0.0")
 PORT = int(os.getenv("PORT", "8000"))
-UVICORN_WORKERS = 8
+UVICORN_WORKERS = 4
 
 logging.basicConfig(level=os.getenv("LOG_LEVEL", "INFO"))
 logger = logging.getLogger("index-service")

search/main.py

@@ -165,7 +165,7 @@ app = FastAPI(title="Search Service", version="0.1.0", lifespan=lifespan)
 DENSE_PREFETCH_K = 80
 SPARSE_PREFETCH_K = 200
 RETRIEVE_K = 150
-RERANK_LIMIT = 10
+RERANK_LIMIT = 60
 
 
 async def embed_dense(client: httpx.AsyncClient, text: str) -> list[float]:
@@ -211,15 +211,16 @@ def embed_sparse_sync(text: str) -> SparseVector:
 
 
 def build_dense_query(question: Question) -> str:
+    if question.search_text:
+        return question.search_text.strip()
     return question.text.strip()
 
 
 def build_sparse_query(question: Question) -> str:
-    parts = [question.text.strip()]
+    base = question.search_text.strip() if question.search_text else question.text.strip()
+    parts = [base]
     if question.keywords:
         parts.extend(question.keywords)
-    if question.search_text:
-        parts = [question.search_text]
     return " ".join(parts)
 
 
@@ -358,13 +359,19 @@ async def search(payload: SearchAPIRequest) -> SearchAPIResponse:
     dense_vector, sparse_vector = await asyncio.gather(dense_task, sparse_task)
 
     dense_vectors = [dense_vector]
-    if question.hyde and len(question.hyde) > 0:
+    extra_texts: list[str] = []
+    for v in (question.variants or [])[:2]:
+        if v.strip():
+            extra_texts.append(v.strip())
+    for h in (question.hyde or [])[:2]:
+        if h.strip():
+            extra_texts.append(h.strip())
+    if extra_texts:
         try:
-            hyde_texts = question.hyde[:2]
-            hyde_vectors = await embed_dense_batch(client, hyde_texts)
-            dense_vectors.extend(hyde_vectors)
+            extra_vectors = await embed_dense_batch(client, extra_texts)
+            dense_vectors.extend(extra_vectors)
         except Exception as e:
-            logger.warning(f"HyDE embedding failed: {e}")
+            logger.warning(f"Extra dense embedding failed: {e}")
 
     all_points = await qdrant_search(qdrant, dense_vectors, sparse_vector)