knowledge_base_examplecsv

# Format: description,Subcategory

# DON'T: add a header row — the parser treats every non-# line as data
# DON'T: use no quotes around the words
# DO: use raw receipt abbreviations as they appear
# DO: use the shortest unique prefix that unambiguously identifies the product category
# tab-separated is accepted in case your spreadsheet exports that way 

# Use the brand name alone when you buy many variants: PARMALAT, BARILLA, MULINO BIANCO
# The RAG system uses semantic embedding similarity, not exact matching — so, for example, `PARMALAT` alone is enough. 
# The embedding model understands that `PARMALAT B12 1.5L` is semantically close to `PARMALAT`. You don't need to enumerate every variant.

# Use the generic Italian noun when brandless: LATTE, PANE, UOVA, POMODORI

BISCOTTI,Snacks
PARMALAT,Dairy & Eggs
PANE,Bakery
LATTE,Dairy & Eggs
MOZZAREL,Dairy & Eggs
POMODOR,Vegetables
BANAN,Fruit
PROSCIUTTO,Meat
SALMONE,Fish & Seafood
LAVATRICE,Cleaning Products
DOVE,Personal Care
BARILLA,Pantry
OLIO,Condiments & Spices
PATATINE,Snacks
LAYS,Snacks
ACQUA NATUR,Beverages
YOGURT,Dairy & Eggs
DIGES. MCVITIE'S,Snacks


# Can DO: include variant spellings and truncations for the same item
MCVITIE,Snacks
MCVITIE'S DIGESTIV,Snacks