Update app.py

app.py

@@ -3,46 +3,41 @@ from huggingface_hub import InferenceClient
 from datasets import load_dataset
 import pandas as pd
 
-# --- Load the Indian Recipes Dataset ---
+# --- Load and Inspect the Indian Recipes Dataset ---
 dataset = load_dataset("nf-analyst/indian_recipe")
 recipes = dataset["train"].to_pandas()
 
-# Preprocess: Combine relevant fields for search
+# Print column names to verify structure
+print("Available columns:", recipes.columns.tolist())
+
+# --- Adjusted Preprocessing ---
+# Based on the dataset's actual columns (replace these with actual column names from the print output)
 recipes["search_text"] = (
-    recipes["name"] + " " + 
-    recipes["ingredients"].apply(lambda x: ' '.join(x)) + " " + 
-    recipes["instructions"].apply(lambda x: ' '.join(x))
+    recipes["RecipeName"] + " " + 
+    recipes["TranslatedIngredients"] + " " + 
+    recipes["TranslatedInstructions"]
 )
 
-# --- Enhanced Search Function (Semantic + Keyword) ---
+# --- Semantic Search Setup ---
 from sentence_transformers import SentenceTransformer
 import faiss
 import numpy as np
 
-# Initialize semantic search
 model = SentenceTransformer("all-MiniLM-L6-v2")
 embeddings = model.encode(recipes["search_text"].tolist())
 index = faiss.IndexFlatL2(embeddings.shape[1])
 index.add(embeddings)
 
 def search_recipes(query, top_k=3):
-    # Semantic search first
     query_embedding = model.encode([query])
     distances, indices = index.search(query_embedding, top_k)
-    
-    # Keyword verification
-    results = []
-    for i in indices[0]:
-        recipe = recipes.iloc[i]
-        if query.lower() in recipe["search_text"].lower():
-            results.append({
-                "name": recipe["name"],
-                "ingredients": recipe["ingredients"],
-                "instructions": recipe["instructions"],
-                "cook_time": recipe["cook_time"],
-                "diet": recipe["diet"]
-            })
-    return results if results else None
+    return [{
+        "name": recipes.iloc[i]["RecipeName"],
+        "ingredients": recipes.iloc[i]["TranslatedIngredients"],
+        "instructions": recipes.iloc[i]["TranslatedInstructions"],
+        "cuisine": recipes.iloc[i]["Cuisine"],
+        "course": recipes.iloc[i]["Course"]
+    } for i in indices[0]]
 
 # --- Modified Respond Function ---
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
@@ -55,36 +50,30 @@ def respond(
     temperature,
     top_p,
 ):
-    # Search our dataset first
     found_recipes = search_recipes(message)
     
     if not found_recipes:
         yield "No matching Indian recipes found. Try terms like 'butter chicken', 'biryani', or 'dal tadka'."
         return
     
-    # Format recipes as strict context
     recipe_context = "\n\n".join([
-        f"Recipe {i+1}: {r['name']} ({r['diet']}, {r['cook_time']})\n"
-        f"Ingredients: {', '.join(r['ingredients'])}\n"
-        f"Method: {' '.join(r['instructions'][:3])}..."
+        f"Recipe {i+1}: {r['name']} ({r['cuisine']}, {r['course']})\n"
+        f"Ingredients: {r['ingredients']}\n"
+        f"Method: {r['instructions'][:200]}..."
         for i, r in enumerate(found_recipes)
     ])
     
-    # Force the LLM to only use these recipes
-    strict_system_prompt = f"""You are an Indian food expert. ONLY recommend from these verified recipes.
-    NEVER invent recipes. If asked for variations, suggest only minor modifications to these:
+    strict_system_prompt = f"""You are an Indian food expert. ONLY recommend from these verified recipes:
     
     {recipe_context}
     
     Respond in this format:
-    1. First recommend recipe names matching the query
-    2. If asked for details, provide ONLY from the recipes above
-    3. For substitutions, suggest similar ingredients from these recipes
-    """
+    1. First list matching recipe names
+    2. Only provide details when explicitly asked
+    3. Never invent recipes or ingredients"""
     
     messages = [{"role": "system", "content": strict_system_prompt}]
     
-    # Add conversation history
     for user_msg, bot_msg in history:
         if user_msg:
             messages.append({"role": "user", "content": user_msg})
@@ -93,7 +82,6 @@ def respond(
     
     messages.append({"role": "user", "content": message})
 
-    # Generate response
     response = ""
     for chunk in client.chat_completion(
         messages,
@@ -106,7 +94,7 @@ def respond(
         response += token
         yield response
 
-# --- Gradio Interface with Indian Food Examples ---
+# --- Gradio Interface ---
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
@@ -121,11 +109,8 @@ demo = gr.ChatInterface(
     examples=[
         "Vegetarian North Indian dinner",
         "Quick chicken curry",
-        "Traditional South Indian breakfast",
-        "Gluten-free Indian dessert"
-    ],
-    title="🍛 Authentic Indian Recipe Assistant",
-    description="Get recommendations ONLY from verified Indian recipes"
+        "Traditional South Indian breakfast"
+    ]
 )
 
 if __name__ == "__main__":