app0.py

import streamlit as st
from openai import OpenAI
from PIL import Image
import io
import base64

st.set_page_config(page_title="Intelligent OCR & Insight", layout="centered", page_icon="🖼️")

st.title("🖼️ Intelligent Image Text Extraction")
st.write("Extract, translate, or summarize text using **meta-llama/llama-3.2-11b-vision-instruct**")

# --- 1. AKILLI API ANAHTARI YÖNETİMİ ---
DEFAULT_API_KEY = "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"

try:
    hf_api_key = st.secrets.get("OPENROUTER_API_KEY", "")
except Exception:
    hf_api_key = ""

# Sol menü (Sidebar) Yapılandırması
with st.sidebar:
    st.header("🔑 API Configuration")
    user_api_key = st.text_input(
        "Custom OpenRouter API Key (Optional)", 
        type="password", 
        help="The system uses a built-in key by default. If the system key reaches its limit, please provide your own key here."
    )
    
    if user_api_key:
        st.success("🎯 Using your custom API key.")
    else:
        st.info("⚡ Using system default API key.")
    
    st.markdown("---")
    st.header("⚙️ Task Settings")
    
    # Seçenekleri ham LLaMA halüsinasyonlarını engelleyecek şekilde optimize ettik
    task_type = st.selectbox(
        "Choose Action",
        ["Direct Text Extraction", "Extract & Translate to Turkish", "Extract & Summarize"]
    )

active_api_key = user_api_key if user_api_key else (hf_api_key if hf_api_key else DEFAULT_API_KEY)

# --- KRİTİK DÜZELTME: LLaMA için kesin sınırlandırılmış promptlar ---
prompt_dict = {
    "Direct Text Extraction": (
        "You are an expert OCR system. Task: Analyze the image and extract all visible text. "
        "Rules: Return ONLY the raw extracted text exactly as it appears in the image. Do not include any explanations, "
        "do not introduce code snippets, do not add introductory phrases like 'Here is the text', and do not output random characters."
    ),
    "Extract & Translate to Turkish": (
    "You are an expert translator. Task: Read the text within the image and translate it line-by-line into fluent Turkish. "
    "Rules: Output ONLY the direct Turkish translation of the document text. Do not summarize it, do not write a description about the letter, "
    "and do not add conversational filler."
    ),
    "Extract & Summarize": (
        "You are an advanced text analytics system. Task: Extract the text from the image and provide a highly structured, "
        "concise summary using bullet points. Rules: Output ONLY the summary. Do not include meta-commentary or technical jargon."
    )
}

# --- 2. GÖRSEL YÜKLEME VE BASE64 ---
uploaded_file = st.file_uploader("Upload an image", type=["png", "jpg", "jpeg"])

def encode_image_to_base64(pil_image):
    buffered = io.BytesIO()
    pil_image.save(buffered, format="PNG")
    return base64.b64encode(buffered.getvalue()).decode('utf-8')

if uploaded_file is not None:
    image = Image.open(uploaded_file)
    st.image(image, caption="Uploaded Image", use_container_width=True)

    if st.button("🔍 Process Image"):
        with st.spinner("LLaMA Vision is analyzing the image..."):
            try:
                client = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=active_api_key)
                base64_image = encode_image_to_base64(image)
                
                response = client.chat.completions.create(
                    model="meta-llama/llama-3.2-11b-vision-instruct",
                    messages=[
                        {
                            "role": "user",
                            "content": [
                                {"type": "text", "text": prompt_dict[task_type]},
                                {
                                    "type": "image_url",
                                    "image_url": {"url": f"data:image/png;base64,{base64_image}"}
                                }
                            ]
                        }
                    ],
                    temperature=0.1 # Düşük sıcaklık: Modelin uydurmasını/halüsinasyon görmesini engeller, netliği artırır!
                )

                extracted_text = response.choices[0].message.content.strip()

                st.subheader("📄 Analysis Result")
                st.text_area("Result Output", extracted_text, height=300)
                
                st.download_button(
                    label="📥 Download Result as .txt",
                    data=extracted_text,
                    file_name="llama_vision_output.txt",
                    mime="text/plain"
                )

            except Exception as e:
                st.error("❌ An error occurred during processing.")
                st.warning(
                    "⚠️ **Quota or Connection Limit Reached:** The built-in shared API key might have run out of credits. "
                    "To continue using the app without interruption, please generate a free/paid key on **OpenRouter.ai** "
                    "and paste it into the **🔑 API Configuration** section in the left sidebar."
                )
                with st.expander("Show technical error details"):
                    st.code(str(e))

# --- SAYFA ALTI (FOOTER) ---
st.markdown("---")
st.markdown("""
### 🚀 Real-World Applications & Use Cases
This vision-language system leveraging **LLaMA 3.2 Vision** goes beyond traditional OCR by understanding the context, layout, and meaning of the text within images.

* **📂 Document Digitization:** Auto-extract tabular data, invoices, and forms directly into structured text.
* **🏥 Healthcare Admin:** Convert scanned doctor prescriptions and reports into digital medical data.
* **🛒 Retail & Logistics:** Process invoices or scan warehouse package labels automatically.
* **🎓 Academic Research:** Convert handwritten study notes or whiteboard layouts into clean text.
* **🏭Intelligent Industrial Data Pipelines & Smart Logistics
* **Context-Aware Quality Control:** Integrates object detection with semantic extraction. For instance, after a vision model identifies an item (e.g., specific produce, automotive part, or electronic kit), this pipeline parses labels to extract critical context like **expiration dates**, **serial numbers**, or **country of origin**.
* **End-to-End Supply Chain Traceability:** Automates compliance checks and inventory logging at shipping hubs without human intervention, reducing friction in dynamic warehouse environments.
""")