app.py

import streamlit as st
from PIL import Image
import torch
import easyocr
import numpy as np
import openai  # Using OpenAI GPT (or replace with GROQ API)
import io
from transformers import CLIPModel, CLIPImageProcessor

# ✅ Fix: set_page_config() must be the first Streamlit command
st.set_page_config(page_title="Multimodal AI Assistant", layout="wide")

# ---- Load CLIP Model (Vision Only) ---- #
@st.cache_resource
def load_clip_model():
    model = CLIPModel.from_pretrained(
        "fxmarty/clip-vision-model-tiny", 
        ignore_mismatched_sizes=True  # ✅ Fix size mismatch
    )
    processor = CLIPImageProcessor.from_pretrained("fxmarty/clip-vision-model-tiny")
    return model, processor

model, processor = load_clip_model()

# ---- Load OCR (EasyOCR) ---- #
@st.cache_resource
def load_ocr():
    return easyocr.Reader(['en'])

reader = load_ocr()

# ---- Streamlit UI ---- #
st.title("🖼️ Multimodal AI Assistant")
st.write("Upload an image, extract text, and ask questions!")

# ---- Upload Image ---- #
uploaded_file = st.file_uploader("📤 Upload an image", type=["jpg", "png", "jpeg"])

extracted_text = None  # Variable to store extracted text

if uploaded_file is not None:
    # Convert file to image format
    image = Image.open(uploaded_file).convert("RGB")

    # ✅ Fix: use `use_container_width` instead of `use_column_width`
    st.image(image, caption="Uploaded Image", use_container_width=True)

    # ✅ Convert PIL image to NumPy array for EasyOCR
    image_np = np.array(image)

    # ✅ Fix: Pass the correct format to EasyOCR
    with st.spinner("🔍 Extracting text from image..."):
        extracted_text_list = reader.readtext(image_np, detail=0)

    extracted_text = " ".join(extracted_text_list)  # Combine extracted text

    st.write("### 📝 Extracted Text:")
    if extracted_text:
        st.success(extracted_text)
    else:
        st.warning("No readable text found in the image.")

# ---- Question Answering Section ---- #
if extracted_text:
    user_question = st.text_input("💡 Ask a question about the extracted text:")

    if user_question:
        with st.spinner("🤖 Thinking..."):
            # Using OpenAI GPT API (replace with GROQ or Hugging Face LLM if needed)
            openai.api_key = "YOUR_OPENAI_API_KEY"  # Store securely in a .env file

            response = openai.ChatCompletion.create(
                model="gpt-3.5-turbo",
                messages=[
                    {"role": "system", "content": "You are an AI assistant helping answer questions based on extracted text from an image."},
                    {"role": "user", "content": f"Extracted text: {extracted_text}\n\nQuestion: {user_question}"}
                ]
            )

            answer = response["choices"][0]["message"]["content"]
        
        st.write("### 🤖 AI Answer:")
        st.success(answer)