Update app.py

app.py

@@ -7,10 +7,9 @@ from PIL import Image
 from transformers import pipeline
 from langchain_chroma import Chroma
 from langchain_community.document_loaders import PyPDFLoader, TextLoader
-from langchain_text_splitters import RecursiveCharacterTextSplitter
 from langchain_core.documents import Document
 from langchain_huggingface import HuggingFaceEmbeddings
-from ultralytics import YOLO
+from ultrultralytics import YOLO
 
 # --- CONFIGURATION ---
 CHROMA_PATH = "/tmp/chroma_db"
@@ -75,7 +74,7 @@ def get_bottle_crops(image_path):
         except:
             return []
 
-# --- RECIPE INGESTION (NOW WITH SCISSORS!) ---
+# --- RECIPE INGESTION (THE "HARD CUT" FIX) ---
 def ingest_recipes(files):
     if not files: return "❌ No files uploaded."
     
@@ -93,27 +92,37 @@ def ingest_recipes(files):
             
     if not docs: return "❌ Could not extract text."
 
-    # --- THE FIX: SPLIT TEXT INTO RECIPES ---
-    # We split by "Recipe:" or newlines to ensure each drink is its own 'chunk'
-    text_splitter = RecursiveCharacterTextSplitter(
-        chunk_size=600,       # Approximate size of one recipe
-        chunk_overlap=50,     # Slight overlap to don't cut words
-        separators=["\nRecipe:", "Recipe:", "\n\n", "\n"] # Priority splitters
-    )
-    splits = text_splitter.split_documents(docs)
-
-    vector_store = Chroma.from_documents(
-        documents=splits, # We ingest the SPLITS, not the whole doc
-        embedding=embed_model, 
-        persist_directory=CHROMA_PATH
-    )
-    return f"✅ Bar library updated. Split into {len(splits)} individual recipes."
+    # 1. Combine all pages/files into one massive text block
+    full_text = "\n".join([d.page_content for d in docs])
+
+    # 2. Strict Split: Cut exactly at the start of any line that says "Recipe:"
+    # (?m)^ means "look at the start of a line"
+    raw_chunks = re.split(r'(?m)^(?=Recipe:)', full_text)
+    
+    split_docs = []
+    for chunk in raw_chunks:
+        # Clean out those long '⸻' separator lines
+        clean_chunk = re.sub(r'⸻+', '', chunk).strip()
+        
+        # If the chunk actually has text in it, save it as a standalone recipe
+        if len(clean_chunk) > 20: 
+            split_docs.append(Document(page_content=clean_chunk))
+
+    # 3. Save to Database
+    try:
+        vector_store = Chroma.from_documents(
+            documents=split_docs, 
+            embedding=embed_model, 
+            persist_directory=CHROMA_PATH
+        )
+        return f"✅ Bar library updated. Strictly split into {len(split_docs)} individual recipes."
+    except Exception as e:
+        return f"❌ Database Error: {e}"
 
 # --- BARTENDER LOGIC ---
 def bartend(message, history, img_path, inventory):
     debug_images = []
     
-    # 1. Vision Scanning
     if img_path:
         crops = get_bottle_crops(img_path)
         debug_images = crops 
@@ -122,7 +131,6 @@ def bartend(message, history, img_path, inventory):
         def identify_spirit(image_input):
             if image_input.mode != "RGB": image_input = image_input.convert("RGB")
             prompt = "User: <image>\nRead the label. What is the specific brand and type of alcohol? Be precise.\nAssistant:"
-            # Positional argument fix
             out = vision_pipe(image_input, prompt, generate_kwargs={"max_new_tokens": 50})
             text = out[0]['generated_text']
             if "Assistant:" in text: return text.split("Assistant:")[-1].strip()
@@ -133,7 +141,6 @@ def bartend(message, history, img_path, inventory):
             inventory = re.sub(r'<.*?>', '', inventory).strip().split('.')[0]
             print(f"🔍 Pass 1 Result: {inventory}")
             
-            # Generic Fallback
             generic_terms = ["vodka", "gin", "rum", "tequila", "whiskey", "whisky", "bourbon", "brandy", "alcohol", "liquor", "spirit", "bottle", "drink"]
             if inventory.lower() in generic_terms or len(inventory) < 4:
                 print("⚠️ Result too generic. Trying FULL IMAGE...")
@@ -147,7 +154,6 @@ def bartend(message, history, img_path, inventory):
             print(f"❌ Vision Failed: {e}")
             inventory = "Unknown Spirit"
 
-    # 2. RAG (Recipe Search)
     recipe_context = ""
     if inventory and inventory not in ["Empty Shelf", "Unknown Spirit", ""]:
         try:
@@ -155,17 +161,16 @@ def bartend(message, history, img_path, inventory):
                 vs = Chroma(persist_directory=CHROMA_PATH, embedding_function=embed_model)
                 search_query = f"Cocktail recipe using {inventory}"
                 
-                # INCREASED K to 5 to give you more options
-                results = vs.similarity_search(search_query, k=5)
-                recipe_context = "\n---\n".join([d.page_content for d in results])
+                # Retrieve the top 4 closest matching recipes
+                results = vs.similarity_search(search_query, k=4)
+                recipe_context = "\n\n---\n\n".join([d.page_content for d in results])
         except Exception as e:
             print(f"Search error: {e}")
 
-    # 3. Create Response
     if inventory == "Unknown Spirit":
         response = "I'm having trouble reading that label. Check the 'Vision Debug' gallery below—is the crop clear?"
     elif recipe_context:
-        response = f"I see you have **{inventory}**. Here are some recipes from your collection:\n\n{recipe_context}"
+        response = f"I see you have **{inventory}**. Here are a few options from your collection:\n\n{recipe_context}"
     else:
         response = f"I see you have **{inventory}**! I don't have a specific recipe for that in the current library. Should I suggest a classic drink?"
 
@@ -188,7 +193,7 @@ with gr.Blocks() as demo:
             gr.Markdown("---")
             img = gr.Image(type="filepath", label="2. Photo of your Bottle")
             
-            with gr.Accordion("🔍 Vision Debug (See what the AI sees)", open=True):
+            with gr.Accordion("🔍 Vision Debug", open=True):
                 debug_gallery = gr.Gallery(label="YOLO Crops", columns=2, height="auto")
             
         with gr.Column(scale=2):