topk query = 50 + 0.55 sim cut off + table chunk size= 2500

index_retriever.py

@@ -52,12 +52,12 @@ def create_query_engine(vector_index):
         vector_retriever = VectorIndexRetriever(
             index=vector_index, 
             similarity_top_k=50,
-            similarity_cutoff=0.65
+            similarity_cutoff=0.55
         )
         
         hybrid_retriever = QueryFusionRetriever(
             [vector_retriever, bm25_retriever],
-            similarity_top_k=40,
+            similarity_top_k=50,
             num_queries=1
         )
         

table_prep.py

@@ -32,7 +32,7 @@ def create_table_content(table_data):
 from llama_index.core.text_splitter import SentenceSplitter
 from config import CHUNK_SIZE, CHUNK_OVERLAP
 
-def chunk_table_document(doc, max_chunk_size=2000):
+def chunk_table_document(doc, max_chunk_size=2500):
     lines = doc.text.strip().split('\n')
 
     # Separate header and data rows