app.py

import gradio as gr
from huggingface_hub import InferenceClient
from datasets import load_dataset
import pandas as pd

# --- Load and Inspect the Indian Recipes Dataset ---
dataset = load_dataset("nf-analyst/indian_recipe")
recipes = dataset["train"].to_pandas()

# 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["RecipeName"] + " " + 
    recipes["TranslatedIngredients"] + " " + 
    recipes["TranslatedInstructions"]
)

# --- Semantic Search Setup ---
from sentence_transformers import SentenceTransformer
import faiss
import numpy as np

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):
    query_embedding = model.encode([query])
    distances, indices = index.search(query_embedding, top_k)
    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")

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    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
    
    recipe_context = "\n\n".join([
        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)
    ])
    
    strict_system_prompt = f"""You are an Indian food expert. ONLY recommend from these verified recipes:
    
    {recipe_context}
    
    Respond in this format:
    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}]
    
    for user_msg, bot_msg in history:
        if user_msg:
            messages.append({"role": "user", "content": user_msg})
        if bot_msg:
            messages.append({"role": "assistant", "content": bot_msg})
    
    messages.append({"role": "user", "content": message})

    response = ""
    for chunk in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = chunk.choices[0].delta.content
        response += token
        yield response

# --- Gradio Interface ---
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(
            value="You are an expert on Indian cuisine.",
            label="System message"
        ),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"),
    ],
    examples=[
        "Vegetarian North Indian dinner",
        "Quick chicken curry",
        "Traditional South Indian breakfast"
    ]
)

if __name__ == "__main__":
    demo.launch()