Create app.py

app.py

@@ -0,0 +1,149 @@
+import gradio as gr
+import pandas as pd
+import pickle
+import torch
+from sentence_transformers import SentenceTransformer, util
+from IO_pipeline import RecipeDigitalizerPipeline  # Importing your image pipeline
+
+# ==========================================
+# 1. LOAD RESOURCES (Run once on startup)
+# ==========================================
+print("⏳ Loading Models and Datasets...")
+
+# A. Load the Sentence Transformer Model (for embedding the NEW recipe)
+model_name = 'BAAI/bge-small-en-v1.5'
+embedding_model = SentenceTransformer(model_name)
+
+# B. Load the Pre-computed Embeddings
+with open('recipe_embeddings.pkl', 'rb') as f:
+    data_bundle = pickle.load(f)
+    # Extract the matrix of vectors (Assuming dict format from previous step)
+    # If you saved just the dataframe, adjust to: stored_embeddings = data_bundle['embedding'].tolist()
+    stored_embeddings = data_bundle['embeddings'] 
+
+# C. Load the CSV Dataset (For displaying recipe details)
+df_recipes = pd.read_csv('RecipeData_10K.csv')
+print("✅ Resources Loaded Successfully!")
+
+
+# ==========================================
+# 2. CORE FUNCTIONS
+# ==========================================
+
+def format_recipe_text(json_data):
+    """Converts the JSON output into a readable string."""
+    if "error" in json_data:
+        return f"Error: {json_data['error']}", ""
+    
+    # Extract fields with safe fallbacks
+    title = json_data.get("title", "Unknown Recipe")
+    cuisine = json_data.get("cuisine_type", "General")
+    difficulty = json_data.get("difficulty", "Medium")
+    
+    ingredients = "\n".join([f"- {item}" for item in json_data.get("ingredients", [])])
+    instructions = "\n".join([f"{i+1}. {step}" for i, step in enumerate(json_data.get("instructions", []))])
+
+    # 1. Readable Text Block
+    display_text = (
+        f"🍽️ RECIPE: {title}\n"
+        f"================================\n"
+        f"🌍 Cuisine: {cuisine}\n"
+        f"⚡ Difficulty: {difficulty}\n\n"
+        f"🛒 INGREDIENTS:\n{ingredients}\n\n"
+        f"🍳 INSTRUCTIONS:\n{instructions}"
+    )
+    
+    # 2. Search Query (Plain text for the AI model)
+    search_query = f"{title} {cuisine} {ingredients} {instructions}"
+    
+    return display_text, search_query
+
+def find_similar_recipes(user_query_text):
+    """Embeds the user's recipe and finds the top 3 matches."""
+    
+    # 1. Embed the new recipe text
+    # BGE model works best with instruction for queries
+    instruction = "Represent this recipe for retrieving similar dishes: "
+    query_embedding = embedding_model.encode(instruction + user_query_text, convert_to_tensor=True)
+    
+    # 2. Compute Cosine Similarity
+    # stored_embeddings must be converted to tensor if it isn't already
+    corpus_embeddings = torch.tensor(stored_embeddings)
+    cos_scores = util.cos_sim(query_embedding, corpus_embeddings)[0]
+    
+    # 3. Get Top 3 Results
+    top_results = torch.topk(cos_scores, k=3)
+    
+    recommendations = ""
+    
+    for score, idx in zip(top_results.values, top_results.indices):
+        idx = int(idx)
+        row = df_recipes.iloc[idx]
+        
+        rec_title = row['Title']
+        # Try to get raw output or construct a summary
+        rec_desc = row['Raw_Output'] if 'Raw_Output' in row else "No description available."
+        # Truncate description for display
+        rec_desc_short = rec_desc[:200] + "..." if len(rec_desc) > 200 else rec_desc
+        
+        recommendations += (
+            f"🏆 MATCH SCORE: {score:.2f}\n"
+            f"📌 {rec_title}\n"
+            f"📝 {rec_desc_short}\n"
+            f"--------------------------------------------------\n"
+        )
+        
+    return recommendations
+
+def magic_pipeline(image_path):
+    # Step 1: Image -> Text (Using your imported IO_pipeline)
+    digitizer = RecipeDigitalizerPipeline()
+    json_result = digitizer.run_pipeline(image_path)
+    
+    # Step 2: Format Text for User
+    readable_text, query_text = format_recipe_text(json_result)
+    
+    # Step 3: Find Similar Recipes (only if we have valid text)
+    if not query_text:
+        return readable_text, "Could not search for similar recipes due to extraction error."
+        
+    similar_recipes_text = find_similar_recipes(query_text)
+    
+    return readable_text, similar_recipes_text
+
+
+# ==========================================
+# 3. GRADIO UI LAYOUT
+# ==========================================
+
+custom_css = """
+#component-0 {max-width: 800px; margin: auto;}
+"""
+
+with gr.Blocks(css=custom_css, theme=gr.themes.Soft()) as demo:
+    
+    gr.Markdown("# 👵 Legacy Kitchen")
+    gr.Markdown("Upload a photo of your handwritten family recipe. We will digitize it and find similar recipes from our database!")
+    
+    with gr.Row():
+        with gr.Column():
+            input_image = gr.Image(type="filepath", label="Upload Recipe Image")
+            submit_btn = gr.Button("✨ Digitize & Find Similar", variant="primary")
+            
+        with gr.Column():
+            # Section 3 & 4: Output Text (Digitized)
+            output_digitized = gr.Textbox(label="📖 Digitized Recipe", lines=10)
+            
+            # Section 5: Similar Recipes Output
+            output_recommendations = gr.Textbox(label="🥗 3 Similar Recipes Found", lines=10)
+
+    # Click Event
+    submit_btn.click(
+        fn=magic_pipeline, 
+        inputs=input_image, 
+        outputs=[output_digitized, output_recommendations]
+    )
+
+# Launch App
+if __name__ == "__main__":
+    demo.launch()