app.py

# ──────────────────────────────── Imports ────────────────────────────────
import os, json, re, logging, requests, markdown, time, io
from datetime import datetime

import streamlit as st
from openai import OpenAI  # OpenAI 라이브러리

from gradio_client import Client
import pandas as pd
import PyPDF2  # For handling PDF files

# ──────────────────────────────── Environment Variables / Constants ─────────────────────────
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "")  
BRAVE_KEY      = os.getenv("SERPHOUSE_API_KEY", "")  # Keep this name
BRAVE_ENDPOINT = "https://api.search.brave.com/res/v1/web/search"
IMAGE_API_URL  = "http://211.233.58.201:7896"
MAX_TOKENS     = 7999

# Blog template and style definitions (in English)
BLOG_TEMPLATES = {
    "ginigen": "Recommended style by Ginigen",
    "standard": "Standard 8-step framework blog",
    "tutorial": "Step-by-step tutorial format",
    "review": "Product/service review format",
    "storytelling": "Storytelling format",
    "seo_optimized": "SEO-optimized blog"
}

BLOG_TONES = {
    "professional": "Professional and formal tone",
    "casual": "Friendly and conversational tone",
    "humorous": "Humorous approach",
    "storytelling": "Story-driven approach"
}

# Example blog topics
EXAMPLE_TOPICS = {
    "example1": "Changes to the real estate tax system in 2025: Impact on average households and tax-saving strategies",
    "example2": "Summer festivals in 2025: A comprehensive guide to major regional events and hidden attractions",
    "example3": "Emerging industries to watch in 2025: An investment guide focused on AI opportunities"
}

# ──────────────────────────────── Logging ────────────────────────────────
logging.basicConfig(level=logging.INFO,
                    format="%(asctime)s - %(levelname)s - %(message)s")

# ──────────────────────────────── OpenAI Client ──────────────────────────

# OpenAI 클라이언트에 타임아웃과 재시도 로직 추가
@st.cache_resource
def get_openai_client():
    """Create an OpenAI client with timeout and retry settings."""
    if not OPENAI_API_KEY:
        raise RuntimeError("⚠️ OPENAI_API_KEY 환경 변수가 설정되지 않았습니다.")
    return OpenAI(
        api_key=OPENAI_API_KEY,
        timeout=60.0,  # 타임아웃 60초로 설정
        max_retries=3  # 재시도 횟수 3회로 설정
    )
    
# ──────────────────────────────── Blog Creation System Prompt ─────────────
def get_system_prompt(template="ginigen", tone="professional", word_count=1750, include_search_results=False, include_uploaded_files=False) -> str:
    """
    Generate a system prompt that includes:
    - The 8-step blog writing framework
    - The selected template and tone
    - Guidelines for using web search results and uploaded files
    """

    # Ginigen recommended style prompt (English version)
    ginigen_prompt = """
You are an expert English SEO copywriter.
◆ Purpose
- Create a blog post based on the given YouTube video script that captivates both search engines and readers.
- Always follow the 4 writing principles: **[Lead with the main point → Keep it simple and short → Emphasize reader benefits → Call to action]**.
◆ Complete Format (Use markdown, avoid unnecessary explanations)
1. **Title**
  - Emoji + Curiosity question/exclamation + Core keywords (Within 70 characters)
  - Example: `# 🧬 Can Reducing Inflammation Help You Lose Weight?! 5 Amazing Benefits of Quercetin`
2. **Hook (2-3 lines)**
  - Present problem → Mention solution keyword → Summarize the benefit of reading this post
3. `---` Divider
4. **Section 1: Core Concept Introduction**
  - `## 🍏 What is [Keyword]?`
  - 1-2 paragraphs definition + 📌 One-line summary
5. `---`
6. **Section 2: 5 Benefits/Reasons**
  - `## 💪 5 Reasons Why [Keyword] Is Beneficial`
  - Each subsection format:
  
### 1. [Keyword-focused subheading]
1-2 paragraphs explanation
> ✔ One-line key point emphasis
- Total of 5 items
7. **Section 3: Consumption/Usage Methods**
- `## 🥗 How to Use [Keyword] Effectively!`
- Emoji bullet list of around 5 items + Additional tips
8. `---`
9. **Concluding Call to Action**
- `## 📌 Conclusion – Start Using [Keyword] Today!`
- 2-3 sentences on benefits/changes → **Action directive** (purchase, subscribe, share, etc.)
10. `---`
11. **Key Summary Table**
 | Item | Effect |
 |---|---|
 | [Keyword] | [Effect summary] |
 | Key foods/products | [List] |
12. `---`
13. **Quiz & CTA**
 - Simple Q&A quiz (1 question) → Reveal answer
 - "If you found this helpful, please share/comment" phrase
 - Preview of next post
◆ Additional Guidelines
- Total length 1,200-1,800 words.
- Use simple vocabulary and short sentences, enhance readability with emojis, bold text, and quoted sections.
- Increase credibility with specific numbers, research results, and analogies.
- No meta-mentions of "prompts" or "instructions".
- Use conversational but professional tone throughout.
- Minimize expressions like "according to research" if no external sources are provided.
◆ Output
- Return **only the completed blog post** in the above format. No additional text.  
"""

    # Standard 8-step framework (English version)
    base_prompt = """
You are an expert in writing professional blog posts. For every blog writing request, strictly follow this 8-step framework to produce a coherent, engaging post:

Reader Connection Phase
1.1. Friendly greeting to build rapport
1.2. Reflect actual reader concerns through introductory questions
1.3. Stimulate immediate interest in the topic

Problem Definition Phase
2.1. Define the reader's pain points in detail
2.2. Analyze the urgency and impact of the problem
2.3. Build a consensus on why it needs to be solved

Establish Expertise Phase
3.1. Analyze based on objective data
3.2. Cite expert views and research findings
3.3. Use real-life examples to further clarify the issue

Solution Phase
4.1. Provide step-by-step guidance
4.2. Suggest practical tips that can be applied immediately
4.3. Mention potential obstacles and how to overcome them

Build Trust Phase
5.1. Present actual success stories
5.2. Quote real user feedback
5.3. Use objective data to prove effectiveness

Action Phase
6.1. Suggest the first clear step the reader can take
6.2. Urge timely action by emphasizing urgency
6.3. Motivate by highlighting incentives or benefits

Authenticity Phase
7.1. Transparently disclose any limits of the solution
7.2. Admit that individual experiences may vary
7.3. Mention prerequisites or cautionary points

Relationship Continuation Phase
8.1. Conclude with sincere gratitude
8.2. Preview upcoming content to build anticipation
8.3. Provide channels for further communication
"""

    # Additional guidelines for each template
    template_guides = {
        "tutorial": """
This blog should be in a tutorial style:
- Clearly state the goal and the final outcome first
- Provide step-by-step explanations with clear separations
- Indicate where images could be inserted for each step
- Mention approximate time requirements and difficulty level
- List necessary tools or prerequisite knowledge
- Give troubleshooting tips and common mistakes to avoid
- Conclude with suggestions for next steps or advanced applications
""",
        "review": """
This blog should be in a review style:
- Separate objective facts from subjective opinions
- Clearly list your evaluation criteria
- Discuss both pros and cons in a balanced way
- Compare with similar products/services
- Specify the target audience for whom it is suitable
- Provide concrete use cases and outcomes
- Conclude with a final recommendation or alternatives
""",
        "storytelling": """
This blog should be in a storytelling style:
- Start with a real or hypothetical person or case
- Emphasize emotional connection with the problem scenario
- Follow a narrative structure centered on conflict and resolution
- Include meaningful insights or lessons learned
- Maintain an emotional thread the reader can relate to
- Balance storytelling with useful information
- Encourage the reader to reflect on their own story
""",
        "seo_optimized": """
This blog should be SEO-optimized:
- Include the main keyword in the title, headings, and first paragraph
- Spread related keywords naturally throughout the text
- Keep paragraphs around 300-500 characters
- Use question-based subheadings
- Make use of lists, tables, and bold text to diversify formatting
- Indicate where internal links could be inserted
- Provide sufficient content of at least 2000-3000 characters
"""
    }

    # Additional guidelines for each tone
    tone_guides = {
        "professional": "Use a professional, authoritative voice. Clearly explain any technical terms and present data or research to maintain a logical flow.",
        "casual": "Use a relaxed, conversational style. Employ personal experiences, relatable examples, and a friendly voice (e.g., 'It's super useful!').",
        "humorous": "Use humor and witty expressions. Add funny analogies or jokes while preserving accuracy and usefulness.",
        "storytelling": "Write as if telling a story, with emotional depth and narrative flow. Incorporate characters, settings, conflicts, and resolutions."
    }

    # Guidelines for using search results
    search_guide = """
Guidelines for Using Search Results:
- Accurately incorporate key information from the search results into the blog
- Include recent data, statistics, and case studies from the search results
- When quoting, specify the source within the text (e.g., "According to XYZ website...")
- At the end of the blog, add a "References" section and list major sources with links
- If there are conflicting pieces of information, present multiple perspectives
- Make sure to reflect the latest trends and data from the search results
"""

    # Guidelines for using uploaded files
    upload_guide = """
Guidelines for Using Uploaded Files (Highest Priority):
- The uploaded files must be a main source of information for the blog
- Carefully examine the data, statistics, or examples in the file and integrate them
- Directly quote and thoroughly explain any key figures or claims from the file
- Highlight the file content as a crucial aspect of the blog
- Mention the source clearly, e.g., "According to the uploaded data..."
- For CSV files, detail important stats or numerical data in the blog
- For PDF files, quote crucial segments or statements
- For text files, integrate relevant content effectively
- Even if the file content seems tangential, do your best to connect it to the blog topic
- Keep consistency throughout and ensure the file's data is appropriately reflected
"""

    # Choose base prompt
    if template == "ginigen":
        final_prompt = ginigen_prompt
    else:
        final_prompt = base_prompt

    # If the user chose a specific template (and not ginigen), append the relevant guidelines
    if template != "ginigen" and template in template_guides:
        final_prompt += "\n" + template_guides[template]

    # If a specific tone is selected, append that guideline
    if tone in tone_guides:
        final_prompt += f"\n\nTone and Manner: {tone_guides[tone]}"

    # If web search results should be included
    if include_search_results:
        final_prompt += f"\n\n{search_guide}"

    # If uploaded files should be included
    if include_uploaded_files:
        final_prompt += f"\n\n{upload_guide}"

    # Word count guidelines
    final_prompt += (
        f"\n\nWriting Requirements:\n"
        f"9.1. Word Count: around {word_count-250}-{word_count+250} characters\n"
        f"9.2. Paragraph Length: 3-4 sentences each\n"
        f"9.3. Visual Cues: Use subheadings, separators, and bullet/numbered lists\n"
        f"9.4. Data: Cite all sources\n"
        f"9.5. Readability: Use clear paragraph breaks and highlights where necessary"
    )

    return final_prompt

# ──────────────────────────────── Brave Search API ────────────────────────
@st.cache_data(ttl=3600)
def brave_search(query: str, count: int = 20):
    """
    Call the Brave Web Search API → list[dict]
    Returns fields: index, title, link, snippet, displayed_link
    """
    if not BRAVE_KEY:
        raise RuntimeError("⚠️ SERPHOUSE_API_KEY (Brave API Key) environment variable is empty.")

    headers = {
        "Accept": "application/json",
        "Accept-Encoding": "gzip",
        "X-Subscription-Token": BRAVE_KEY
    }
    params = {"q": query, "count": str(count)}

    for attempt in range(3):
        try:
            r = requests.get(BRAVE_ENDPOINT, headers=headers, params=params, timeout=15)
            r.raise_for_status()
            data = r.json()

            logging.info(f"Brave search result data structure: {list(data.keys())}")

            raw = data.get("web", {}).get("results") or data.get("results", [])
            if not raw:
                logging.warning(f"No Brave search results found. Response: {data}")
                raise ValueError("No search results found.")
            
            arts = []
            for i, res in enumerate(raw[:count], 1):
                url = res.get("url", res.get("link", ""))
                host = re.sub(r"https?://(www\.)?", "", url).split("/")[0]
                arts.append({
                    "index": i,
                    "title": res.get("title", "No title"),
                    "link": url,
                    "snippet": res.get("description", res.get("text", "No snippet")),
                    "displayed_link": host
                })

            logging.info(f"Brave search success: {len(arts)} results")
            return arts

        except Exception as e:
            logging.error(f"Brave search failure (attempt {attempt+1}/3): {e}")
            if attempt < 2:
                time.sleep(2)

    return []

def mock_results(query: str) -> str:
    """Fallback search results if API fails"""
    ts = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
    return (f"# Fallback Search Content (Generated: {ts})\n\n"
            f"The search API request failed. Please generate the blog based on any pre-existing knowledge about '{query}'.\n\n"
            f"You may consider the following points:\n\n"
            f"- Basic concepts and importance of {query}\n"
            f"- Commonly known related statistics or trends\n"
            f"- Typical expert opinions on this subject\n"
            f"- Questions that readers might have\n\n"
            f"Note: This is fallback guidance, not real-time data.\n\n")

def do_web_search(query: str) -> str:
    """Perform web search and format the results."""
    try:
        arts = brave_search(query, 20)
        if not arts:
            logging.warning("No search results, using fallback content")
            return mock_results(query)

        hdr = "# Web Search Results\nUse the information below to enhance the reliability of your blog. When you quote, please cite the source, and add a References section at the end of the blog.\n\n"
        body = "\n".join(
            f"### Result {a['index']}: {a['title']}\n\n{a['snippet']}\n\n"
            f"**Source**: [{a['displayed_link']}]({a['link']})\n\n---\n"
            for a in arts
        )
        return hdr + body
    except Exception as e:
        logging.error(f"Web search process failed: {str(e)}")
        return mock_results(query)

# ──────────────────────────────── File Upload Handling ─────────────────────
def process_text_file(file):
    """Handle text file"""
    try:
        content = file.read()
        file.seek(0)

        text = content.decode('utf-8', errors='ignore')
        if len(text) > 10000:
            text = text[:9700] + "...(truncated)..."

        result = f"## Text File: {file.name}\n\n"
        result += text
        return result
    except Exception as e:
        logging.error(f"Error processing text file: {str(e)}")
        return f"Error processing text file: {str(e)}"

def process_csv_file(file):
    """Handle CSV file"""
    try:
        content = file.read()
        file.seek(0)

        df = pd.read_csv(io.BytesIO(content))
        result = f"## CSV File: {file.name}\n\n"
        result += f"- Rows: {len(df)}\n"
        result += f"- Columns: {len(df.columns)}\n"
        result += f"- Column Names: {', '.join(df.columns.tolist())}\n\n"

        result += "### Data Preview\n\n"
        preview_df = df.head(10)
        try:
            markdown_table = preview_df.to_markdown(index=False)
            if markdown_table:
                result += markdown_table + "\n\n"
            else:
                result += "Unable to display CSV data.\n\n"
        except Exception as e:
            logging.error(f"Markdown table conversion error: {e}")
            result += "Displaying data as text:\n\n"
            result += str(preview_df) + "\n\n"

        num_cols = df.select_dtypes(include=['number']).columns
        if len(num_cols) > 0:
            result += "### Basic Statistical Information\n\n"
            try:
                stats_df = df[num_cols].describe().round(2)
                stats_markdown = stats_df.to_markdown()
                if stats_markdown:
                    result += stats_markdown + "\n\n"
                else:
                    result += "Unable to display statistical information.\n\n"
            except Exception as e:
                logging.error(f"Statistical info conversion error: {e}")
                result += "Unable to generate statistical information.\n\n"

        return result
    except Exception as e:
        logging.error(f"CSV file processing error: {str(e)}")
        return f"Error processing CSV file: {str(e)}"

def process_pdf_file(file):
    """Handle PDF file"""
    try:
        # Read file in bytes
        file_bytes = file.read()
        file.seek(0)

        # Use PyPDF2
        pdf_file = io.BytesIO(file_bytes)
        reader = PyPDF2.PdfReader(pdf_file, strict=False)

        # Basic info
        result = f"## PDF File: {file.name}\n\n"
        result += f"- Total pages: {len(reader.pages)}\n\n"

        # Extract text by page (limit to first 5 pages)
        max_pages = min(5, len(reader.pages))
        all_text = ""

        for i in range(max_pages):
            try:
                page = reader.pages[i]
                page_text = page.extract_text()

                current_page_text = f"### Page {i+1}\n\n"
                if page_text and len(page_text.strip()) > 0:
                    # Limit to 1500 characters per page
                    if len(page_text) > 1500:
                        current_page_text += page_text[:1500] + "...(truncated)...\n\n"
                    else:
                        current_page_text += page_text + "\n\n"
                else:
                    current_page_text += "(No text could be extracted from this page)\n\n"

                all_text += current_page_text

                # If total text is too long, break
                if len(all_text) > 8000:
                    all_text += "...(truncating remaining pages; PDF is too large)...\n\n"
                    break

            except Exception as page_err:
                logging.error(f"Error processing PDF page {i+1}: {str(page_err)}")
                all_text += f"### Page {i+1}\n\n(Error extracting content: {str(page_err)})\n\n"

        if len(reader.pages) > max_pages:
            all_text += f"\nNote: Only the first {max_pages} pages are shown out of {len(reader.pages)} total.\n\n"

        result += "### PDF Content\n\n" + all_text
        return result

    except Exception as e:
        logging.error(f"PDF file processing error: {str(e)}")
        return f"## PDF File: {file.name}\n\nError occurred: {str(e)}\n\nThis PDF file cannot be processed."

def process_uploaded_files(files):
    """Combine the contents of all uploaded files into one string."""
    if not files:
        return None

    result = "# Uploaded File Contents\n\n"
    result += "Below is the content from the files provided by the user. Integrate this data as a main source of information for the blog.\n\n"

    for file in files:
        try:
            ext = file.name.split('.')[-1].lower()
            if ext == 'txt':
                result += process_text_file(file) + "\n\n---\n\n"
            elif ext == 'csv':
                result += process_csv_file(file) + "\n\n---\n\n"
            elif ext == 'pdf':
                result += process_pdf_file(file) + "\n\n---\n\n"
            else:
                result += f"### Unsupported File: {file.name}\n\n---\n\n"
        except Exception as e:
            logging.error(f"File processing error {file.name}: {e}")
            result += f"### File processing error: {file.name}\n\nError: {e}\n\n---\n\n"

    return result

# ──────────────────────────────── Image & Utility ─────────────────────────
def generate_image(prompt, w=768, h=768, g=3.5, steps=30, seed=3):
    """Image generation function."""
    if not prompt:
        return None, "Insufficient prompt"
    try:
        res = Client(IMAGE_API_URL).predict(
            prompt=prompt, width=w, height=h, guidance=g,
            inference_steps=steps, seed=seed,
            do_img2img=False, init_image=None,
            image2image_strength=0.8, resize_img=True,
            api_name="/generate_image"
        )
        return res[0], f"Seed: {res[1]}"
    except Exception as e:
        logging.error(e)
        return None, str(e)

def extract_image_prompt(blog_text: str, topic: str):
    """
    Generate a single-line English image prompt from the blog content.
    """
    client = get_openai_client()
    
    try:
        response = client.chat.completions.create(
            model="gpt-4.1-mini",  # 일반적으로 사용 가능한 모델로 설정
            messages=[
                {"role": "system", "content": "Generate a single-line English image prompt from the following text. Return only the prompt text, nothing else."},
                {"role": "user", "content": f"Topic: {topic}\n\n---\n{blog_text}\n\n---"}
            ],
            temperature=1,
            max_tokens=80,
            top_p=1
        )
        
        return response.choices[0].message.content.strip()
    except Exception as e:
        logging.error(f"OpenAI image prompt generation error: {e}")
        return f"A professional photo related to {topic}, high quality"

def md_to_html(md: str, title="Ginigen Blog"):
    """Convert Markdown to HTML."""
    return f"<!DOCTYPE html><html><head><title>{title}</title><meta charset='utf-8'></head><body>{markdown.markdown(md)}</body></html>"

def keywords(text: str, top=5):
    """Simple keyword extraction."""
    cleaned = re.sub(r"[^가-힣a-zA-Z0-9\s]", "", text)
    return " ".join(cleaned.split()[:top])

# ──────────────────────────────── Streamlit UI ────────────────────────────
def ginigen_app():
    st.title("Ginigen Blog")

    # Set default session state
    if "ai_model" not in st.session_state:
        st.session_state.ai_model = "gpt-4.1-mini"  # 고정 모델 설정
    if "messages" not in st.session_state:
        st.session_state.messages = []
    if "auto_save" not in st.session_state:
        st.session_state.auto_save = True
    if "generate_image" not in st.session_state:
        st.session_state.generate_image = False
    if "web_search_enabled" not in st.session_state:
        st.session_state.web_search_enabled = True
    if "blog_template" not in st.session_state:
        st.session_state.blog_template = "ginigen"  # Ginigen recommended style by default
    if "blog_tone" not in st.session_state:
        st.session_state.blog_tone = "professional"
    if "word_count" not in st.session_state:
        st.session_state.word_count = 1750

    # Sidebar UI
    sb = st.sidebar
    sb.title("Blog Settings")
    
    # 모델 선택 제거 (고정 모델 사용)
    
    sb.subheader("Blog Style Settings")
    sb.selectbox(
        "Blog Template", 
        options=list(BLOG_TEMPLATES.keys()), 
        format_func=lambda x: BLOG_TEMPLATES[x],
        key="blog_template"
    )
    
    sb.selectbox(
        "Blog Tone",
        options=list(BLOG_TONES.keys()),
        format_func=lambda x: BLOG_TONES[x],
        key="blog_tone"
    )
    
    sb.slider("Blog Length (word count)", 800, 3000, key="word_count")

    
    # Example topics
    sb.subheader("Example Topics")
    c1, c2, c3 = sb.columns(3)
    if c1.button("Real Estate Tax", key="ex1"):
        process_example(EXAMPLE_TOPICS["example1"])
    if c2.button("Summer Festivals", key="ex2"):
        process_example(EXAMPLE_TOPICS["example2"])
    if c3.button("Investment Guide", key="ex3"):
        process_example(EXAMPLE_TOPICS["example3"])
    
    sb.subheader("Other Settings")
    sb.toggle("Auto Save", key="auto_save")
    sb.toggle("Auto Image Generation", key="generate_image")
    
    web_search_enabled = sb.toggle("Use Web Search", value=st.session_state.web_search_enabled)
    st.session_state.web_search_enabled = web_search_enabled
    
    if web_search_enabled:
        st.sidebar.info("✅ Web search results will be integrated into the blog.")

    # Download the latest blog (markdown/HTML)
    latest_blog = next(
        (m["content"] for m in reversed(st.session_state.messages) 
         if m["role"] == "assistant" and m["content"].strip()), 
        None
    )
    if latest_blog:
        title_match = re.search(r"# (.*?)(\n|$)", latest_blog)
        title = title_match.group(1).strip() if title_match else "blog"
        sb.subheader("Download Latest Blog")
        d1, d2 = sb.columns(2)
        d1.download_button("Download as Markdown", latest_blog, 
                           file_name=f"{title}.md", mime="text/markdown")
        d2.download_button("Download as HTML", md_to_html(latest_blog, title),
                           file_name=f"{title}.html", mime="text/html")

    # JSON conversation record upload
    up = sb.file_uploader("Load Conversation History (.json)", type=["json"], key="json_uploader")
    if up:
        try:
            st.session_state.messages = json.load(up)
            sb.success("Conversation history loaded successfully")
        except Exception as e:
            sb.error(f"Failed to load: {e}")

    # JSON conversation record download
    if sb.button("Download Conversation as JSON"):
        sb.download_button(
            "Save",
            data=json.dumps(st.session_state.messages, ensure_ascii=False, indent=2),
            file_name="chat_history.json",
            mime="application/json"
        )

    # File Upload
    st.subheader("File Upload")
    uploaded_files = st.file_uploader(
        "Upload files to be referenced in your blog (txt, csv, pdf)",
        type=["txt", "csv", "pdf"],
        accept_multiple_files=True,
        key="file_uploader"
    )
    
    if uploaded_files:
        file_count = len(uploaded_files)
        st.success(f"{file_count} files uploaded. They will be referenced in the blog.")
        
        with st.expander("Preview Uploaded Files", expanded=False):
            for idx, file in enumerate(uploaded_files):
                st.write(f"**File Name:** {file.name}")
                ext = file.name.split('.')[-1].lower()
                
                if ext == 'txt':
                    preview = file.read(1000).decode('utf-8', errors='ignore')
                    file.seek(0)
                    st.text_area(
                        f"Preview of {file.name}",
                        preview + ("..." if len(preview) >= 1000 else ""),
                        height=150
                    )
                elif ext == 'csv':
                    try:
                        df = pd.read_csv(file)
                        file.seek(0)
                        st.write("CSV Preview (up to 5 rows)")
                        st.dataframe(df.head(5))
                    except Exception as e:
                        st.error(f"CSV preview failed: {e}")
                elif ext == 'pdf':
                    try:
                        file_bytes = file.read()
                        file.seek(0)
                        
                        pdf_file = io.BytesIO(file_bytes)
                        reader = PyPDF2.PdfReader(pdf_file, strict=False)
                        
                        pc = len(reader.pages)
                        st.write(f"PDF File: {pc} pages")
                        
                        if pc > 0:
                            try:
                                page_text = reader.pages[0].extract_text()
                                preview = page_text[:500] if page_text else "(No text extracted)"
                                st.text_area("Preview of the first page", preview + "...", height=150)
                            except:
                                st.warning("Failed to extract text from the first page")
                    except Exception as e:
                        st.error(f"PDF preview failed: {e}")

                if idx < file_count - 1:
                    st.divider()

    # Display existing messages
    for m in st.session_state.messages:
        with st.chat_message(m["role"]):
            st.markdown(m["content"])
            if "image" in m:
                st.image(m["image"], caption=m.get("image_caption", ""))

    # User input
    prompt = st.chat_input("Enter a blog topic or keywords.")
    if prompt:
        process_input(prompt, uploaded_files)

def process_example(topic):
    """Process the selected example topic."""
    process_input(topic, [])

def process_input(prompt: str, uploaded_files):
    # Add user's message
    if not any(m["role"] == "user" and m["content"] == prompt for m in st.session_state.messages):
        st.session_state.messages.append({"role": "user", "content": prompt})

    with st.chat_message("user"):
        st.markdown(prompt)
    
    with st.chat_message("assistant"):
        placeholder = st.empty()
        message_placeholder = st.empty()
        full_response = ""

        use_web_search = st.session_state.web_search_enabled
        has_uploaded_files = bool(uploaded_files) and len(uploaded_files) > 0
        
        try:
            # 상태 표시를 위한 상태 컴포넌트
            status = st.status("Preparing to generate blog...")
            status.update(label="Initializing client...")
            
            client = get_openai_client()
            
            # Prepare conversation messages
            messages = []
            
            # Web search
            search_content = None
            if use_web_search:
                status.update(label="Performing web search...")
                with st.spinner("Searching the web..."):
                    search_content = do_web_search(keywords(prompt, top=5))
            
            # Process uploaded files → content
            file_content = None
            if has_uploaded_files:
                status.update(label="Processing uploaded files...")
                with st.spinner("Analyzing files..."):
                    file_content = process_uploaded_files(uploaded_files)
            
            # Build system prompt
            status.update(label="Preparing blog draft...")
            sys_prompt = get_system_prompt(
                template=st.session_state.blog_template,
                tone=st.session_state.blog_tone,
                word_count=st.session_state.word_count,
                include_search_results=use_web_search,
                include_uploaded_files=has_uploaded_files
            )

            # OpenAI API 호출 준비
            status.update(label="Writing blog content...")
            
            # 메시지 구성
            api_messages = [
                {"role": "system", "content": sys_prompt}
            ]
            
            user_content = prompt
            
            # 검색 결과가 있으면 사용자 프롬프트에 추가
            if search_content:
                user_content += "\n\n" + search_content
            
            # 파일 내용이 있으면 사용자 프롬프트에 추가
            if file_content:
                user_content += "\n\n" + file_content
            
            # 사용자 메시지 추가
            api_messages.append({"role": "user", "content": user_content})
            
            # OpenAI API 스트리밍 호출 - 고정 모델 "gpt-4.1-mini" 사용
            try:
                # 스트리밍 방식으로 API 호출
                stream = client.chat.completions.create(
                    model="gpt-4.1-mini",  # 고정 모델 사용
                    messages=api_messages,
                    temperature=1,
                    max_tokens=MAX_TOKENS,
                    top_p=1,
                    stream=True  # 스트리밍 활성화
                )
                
                # 스트리밍 응답 처리
                for chunk in stream:
                    if chunk.choices and len(chunk.choices) > 0 and chunk.choices[0].delta.content is not None:
                        content_delta = chunk.choices[0].delta.content
                        full_response += content_delta
                        message_placeholder.markdown(full_response + "▌")
                
                # 최종 응답 표시 (커서 제거)
                message_placeholder.markdown(full_response)
                status.update(label="Blog completed!", state="complete")
                
            except Exception as api_error:
                error_message = str(api_error)
                logging.error(f"API error: {error_message}")
                status.update(label=f"Error: {error_message}", state="error")
                raise Exception(f"Blog generation error: {error_message}")
            
            # 이미지 생성
            answer_entry_saved = False
            if st.session_state.generate_image and full_response:
                with st.spinner("Generating image..."):
                    try:
                        ip = extract_image_prompt(full_response, prompt)
                        img, cap = generate_image(ip)
                        if img:
                            st.image(img, caption=cap)
                            st.session_state.messages.append({
                                "role": "assistant",
                                "content": full_response,
                                "image": img,
                                "image_caption": cap
                            })
                            answer_entry_saved = True
                    except Exception as img_error:
                        logging.error(f"Image generation error: {str(img_error)}")
                        st.warning("이미지 생성에 실패했습니다. 블로그 콘텐츠만 저장됩니다.")

            # Save the answer if not saved above
            if not answer_entry_saved and full_response:
                st.session_state.messages.append({"role": "assistant", "content": full_response})

            # Download buttons
            if full_response:
                st.subheader("Download This Blog")
                c1, c2 = st.columns(2)
                c1.download_button(
                    "Markdown", 
                    data=full_response, 
                    file_name=f"{prompt[:30]}.md",
                    mime="text/markdown"
                )
                c2.download_button(
                    "HTML",
                    data=md_to_html(full_response, prompt[:30]),
                    file_name=f"{prompt[:30]}.html",
                    mime="text/html"
                )

            # Auto save
            if st.session_state.auto_save and st.session_state.messages:
                try:
                    fn = f"chat_history_auto_{datetime.now():%Y%m%d_%H%M%S}.json"
                    with open(fn, "w", encoding="utf-8") as fp:
                        json.dump(st.session_state.messages, fp, ensure_ascii=False, indent=2)
                except Exception as e:
                    logging.error(f"Auto-save failed: {e}")

        except Exception as e:
            error_message = str(e)
            placeholder.error(f"An error occurred: {error_message}")
            logging.error(f"Process input error: {error_message}")
            ans = f"An error occurred while processing your request: {error_message}"
            st.session_state.messages.append({"role": "assistant", "content": ans})


# ──────────────────────────────── main ────────────────────────────────────
def main():
    ginigen_app()

if __name__ == "__main__":
    main()