top k reranker = 20, max rows = 10, max chars= 2000 + new deduplication

top k reranker = 20, max rows = 10, max chars= 4000 + new deduplication

top k reranker = 25, max rows = 5, max chars= 4000

Much lower reranking threshold (-0.5 instead of 0.1) + detailed score logging

removed normalization doc id

removed normalization doc id

index retriever = 100 + 100

index retriever = 100 + 100

max_chars = 1500 + doc id retriever

new keyword score based index retriever + answer question

new keyword score based index retriever + answer question

chunk size = 1024 + max chars = 1200 + deduplication variant

chunk siz = 1000, max_chars = 1500

new embeeding model + new create_quer_engine with keyword matching

max chars = 2000 for tables + new answer_question

max rows = 10 + new answer_question + reranking

removed query enhanced

simplest version

simplest version

simplest version

simplest version

simplest version

simplest version

simplest version

simplest version

simplest version

simplest version

simplest version

table prep changed

hybrid_retrieve_with_keywords implemented

a new way with keywords

bm = 20, semantic = 30. hybrid = 40, cutoff = 0.65, topk_reranker= 20

bm = 30, semantic = 30. hybrid = 60, cutoff = 0.6, topk_reranker= 40

bm = 30, semantic = 30. hybrid = 60, cutoff = 0.7, topk_reranker= 60

bm = 40, semantic = 20. hybrid = 50, cutoff = 0.7, topk_reranker= 40

bm = 45, semantic = 20. hybrid = 50, cutoff = 0.7, topk_reranker= 20

top k in reranker = 30, bm = 20

bm = 50, semantic = 50. hybrid = 50

bm = 40, semantic = 50. hybrid = 40

bm = 25, semantic = 35. hybrid = 40

with bm and semantic

only semantic search top k = 30, cut off = 0.7

added 35 semantic similarity

new ways

new ways

new ways

a new way = sim cutoff = 0,65

a new way = sim cutoff = 0,65

new way of retrieving

new way of retrieving