app.py

import gradio as gr
import os
import json
import re
from typing import Iterator, Dict, Any, List, Optional
from openai import OpenAI
from openai.types.chat import ChatCompletionChunk

# Load abstracts content once at startup
def load_abstracts_content():
    """Load the abstracts content once at startup to avoid reading file on every request."""
    try:
        with open("abstracts.md", "r", encoding="utf-8") as f:
            return f.read()
    except FileNotFoundError:
        return "Abstracts database not found."

# Load abstracts content globally
ABSTRACTS_CONTENT = load_abstracts_content()

# Load full paper texts
def load_paper_texts():
    """Load all paper texts from the Papers directory and create a mapping from abstracts filenames."""
    papers = {}
    papers_dir = "Papers"
    
    if not os.path.exists(papers_dir):
        return {}
    
    # Create a mapping from abstracts filenames to actual file content
    for filename in os.listdir(papers_dir):
        if filename.endswith('.txt'):
            filepath = os.path.join(papers_dir, filename)
            try:
                with open(filepath, "r", encoding="utf-8") as f:
                    content = f.read()
                    # Store with the filename as key
                    papers[filename] = content
            except Exception as e:
                papers[filename] = f"Error loading paper: {str(e)}"
    
    return papers

# Load paper texts globally
PAPER_TEXTS = load_paper_texts()

def normalize_filename(filename):
    """Normalize filename for better matching."""
    # Remove .txt extension and normalize
    if filename.endswith('.txt'):
        filename = filename[:-4]
    
    # Convert to lowercase and remove special characters
    filename = re.sub(r'[^\w\s]', '', filename.lower())
    # Normalize whitespace
    filename = ' '.join(filename.split())
    return filename

def find_matching_paper_file(query_terms, papers_dict):
    """Find the best matching paper file based on query terms."""
    query_normalized = normalize_filename(' '.join(query_terms))
    best_match = None
    best_score = 0
    
    for filename in papers_dict.keys():
        filename_normalized = normalize_filename(filename)
        
        # Calculate match score
        score = 0
        
        # Exact substring match
        if query_normalized in filename_normalized or filename_normalized in query_normalized:
            score += 10
        
        # Word overlap
        query_words = set(query_normalized.split())
        filename_words = set(filename_normalized.split())
        overlap = len(query_words.intersection(filename_words))
        score += overlap * 2
        
        # Partial word matches
        for query_word in query_words:
            for filename_word in filename_words:
                if query_word in filename_word or filename_word in query_word:
                    score += 1
        
        if score > best_score:
            best_score = score
            best_match = filename
    
    return best_match if best_score > 0 else None

def get_relevant_papers_content(query, max_papers=5):
    """Get relevant paper content based on user query."""
    query_terms = query.lower().split()
    relevant_papers = []
    
    for filename, content in PAPER_TEXTS.items():
        title = filename[:-4] if filename.endswith('.txt') else filename
        title_lower = title.lower()
        
        # Calculate relevance score
        score = 0
        for term in query_terms:
            if term in title_lower:
                score += 2
            if term in content.lower():
                score += 1
        
        if score > 0:
            relevant_papers.append((filename, content, score))
    
    # Sort by relevance score and return top papers
    relevant_papers.sort(key=lambda x: x[2], reverse=True)
    return relevant_papers[:max_papers]

def get_full_paper_content(title, max_chars=12000):
    """Get full paper content for a specific title."""
    for filename, content in PAPER_TEXTS.items():
        if title.lower() in filename.lower() or filename.lower() in title.lower():
            return content[:max_chars] + "..." if len(content) > max_chars else content
    return "Paper not found."

def get_paper_summary(title):
    """Get a structured summary of a paper."""
    content = get_full_paper_content(title)
    if content == "Paper not found.":
        return content
    
    # Extract key sections
    sections = {
        'abstract': '',
        'introduction': '',
        'methodology': '',
        'results': '',
        'conclusions': ''
    }
    
    lines = content.split('\n')
    current_section = None
    
    for line in lines:
        line_lower = line.lower().strip()
        
        # Detect section headers
        if any(keyword in line_lower for keyword in ['abstract', 'introduction', 'method', 'methodology', 'results', 'conclusion']):
            if 'abstract' in line_lower:
                current_section = 'abstract'
            elif 'introduction' in line_lower:
                current_section = 'introduction'
            elif 'method' in line_lower:
                current_section = 'methodology'
            elif 'result' in line_lower:
                current_section = 'results'
            elif 'conclusion' in line_lower:
                current_section = 'conclusions'
        
        # Add content to current section
        if current_section and line.strip():
            sections[current_section] += line + '\n'
    
    # Create structured summary
    summary = f"# {title}\n\n"
    for section, content in sections.items():
        if content.strip():
            summary += f"## {section.title()}\n{content.strip()}\n\n"
    
    return summary

# Get API key with better error handling
api_key = os.getenv("OPENAI_API_KEY")
if not api_key:
    print("⚠️  Warning: OPENAI_API_KEY environment variable not set!")
    client = None
else:
    client = OpenAI(
        api_key=api_key,
        timeout=60.0,
        max_retries=3
    )

# Available models
AVAILABLE_MODELS = {
    "GPT-4o-mini": "gpt-4o-mini",
    "GPT-4o": "gpt-4o",
    "GPT-3.5 Turbo": "gpt-3.5-turbo"
}

# Define the tool for fetching papers
FETCH_PAPERS_TOOL = {
    "type": "function",
    "function": {
        "name": "fetch_papers",
        "description": "Fetch full text content of research papers by their filenames. Use this when you need detailed information, full text, conclusions, methodology, or specific quotes from papers.",
        "parameters": {
            "type": "object",
            "properties": {
                "filenames": {
                    "type": "array",
                    "items": {
                        "type": "string"
                    },
                    "description": "List of paper filenames to fetch (e.g., ['The Labor Market Effects of Generativ.txt', 'AI Companions Reduce Loneliness.txt'])"
                }
            },
            "required": ["filenames"]
        }
    }
}

def fetch_papers(filenames: List[str]) -> Dict[str, str]:
    """
    Fetch full paper texts by filenames.
    Returns a dictionary mapping filename to content.
    """
    papers = {}
    papers_dir = "Papers"
    
    if not os.path.exists(papers_dir):
        return {"error": "Papers directory not found"}
    
    for filename in filenames:
        # Ensure .txt extension
        if not filename.endswith('.txt'):
            filename += '.txt'
            
        filepath = os.path.join(papers_dir, filename)
        
        if os.path.exists(filepath):
            try:
                with open(filepath, "r", encoding="utf-8") as f:
                    papers[filename] = f.read()
            except Exception as e:
                papers[filename] = f"Error loading paper: {str(e)}"
        else:
            papers[filename] = f"Paper not found: {filename}"
    
    return papers

def extract_conclusion_from_paper(content: str) -> str:
    """Extract the conclusion section from a paper's content."""
    conclusion_patterns = [
        "conclusion and future works",
        "conclusion and future work", 
        "conclusions",
        "conclusion",
        "summary and conclusions",
        "discussion and conclusions"
    ]
    
    lines = content.split('\n')
    conclusion_start = -1
    
    for i, line in enumerate(lines):
        line_lower = line.lower().strip()
        if any(pattern in line_lower for pattern in conclusion_patterns):
            if (line.isupper() or 
                line.strip().endswith(':') or 
                len(line.strip()) < 100 or
                line.strip().startswith('Conclusion')):
                conclusion_start = i
                break
    
    if conclusion_start != -1:
        conclusion_lines = []
        for line in lines[conclusion_start:]:
            line_stripped = line.strip()
            if (line_stripped.lower().startswith('acknowledgments') or 
                line_stripped.lower().startswith('references') or
                line_stripped.startswith('--- Page')):
                break
            conclusion_lines.append(line)
        
        return '\n'.join(conclusion_lines)
    
    # Fallback: return the last 1000 characters
    return content[-1000:] if len(content) > 1000 else content

def truncate_conversation_history(messages: list, max_tokens: int = 8000) -> list:
    """Truncate conversation history to stay within token limits."""
    if len(messages) <= 3:
        return messages
    
    system_message = messages[0]
    conversation_messages = messages[1:]
    
    while len(conversation_messages) > 6:
        conversation_messages = conversation_messages[2:]
    
    return [system_message] + conversation_messages

def respond(
    message: str,
    history: list[tuple[str, str]],
    model_name: str,
    max_tokens: int,
    temperature: float,
    top_p: float,
) -> Iterator[str]:
    """
    Generate a response using OpenAI's models with function calling.
    """
    if not client:
        yield "❌ Error: OpenAI API key not configured."
        return
    
    if not message.strip():
        yield "Please enter a message to start the conversation."
        return
    
    # Get relevant full paper content based on user query
    relevant_papers_content = get_relevant_papers_content(message)

    # Check if user is asking for a specific paper (e.g., "show me the full paper about pigs")
    specific_paper_content = ""
    conclusion_content = ""
    paper_summary_content = ""

    if any(keyword in message.lower() for keyword in ["full paper", "complete paper", "entire paper", "show me the paper", "read the paper", "summarize", "summary"]):
        # Try to find specific paper content
        for filename, content in PAPER_TEXTS.items():
            title = filename[:-4] if filename.endswith('.txt') else filename
            if any(term in title.lower() for term in message.lower().split()):
                if any(keyword in message.lower() for keyword in ["summarize", "summary"]):
                    paper_summary_content = get_paper_summary(title)
                else:
                    specific_paper_content = get_full_paper_content(title)
                break

    # Check if user is asking for conclusions specifically
    if any(keyword in message.lower() for keyword in ["conclusion", "conclusions", "what's the conclusion", "what is the conclusion"]):
        for filename, content in PAPER_TEXTS.items():
            title = filename[:-4] if filename.endswith('.txt') else filename
            if any(term in title.lower() for term in message.lower().split()):
                conclusion_content = extract_conclusion_from_paper(content)
                break

    # Initialize messages with a comprehensive system prompt
    system_prompt = f"""You are an AI chatbot designed to help users explore and analyze AI research papers. 

You have access to:
1. An abstracts database with summaries of research papers
2. Full paper texts for detailed analysis
3. A tool to fetch additional paper content when needed

ABSTRACTS DATABASE:
{ABSTRACTS_CONTENT}

RELEVANT PAPERS CONTENT:
{chr(10).join([f"Paper: {filename}{chr(10)}Content: {content[:3000]}..." for filename, content, score in relevant_papers_content])}

SPECIFIC PAPER CONTENT:
{specific_paper_content if specific_paper_content else "None"}

CONCLUSION CONTENT:
{conclusion_content if conclusion_content else "None"}

PAPER SUMMARY:
{paper_summary_content if paper_summary_content else "None"}

INSTRUCTIONS:
- Use the abstracts for general questions and overview
- Use full paper content when users ask for specific details, conclusions, or complete papers
- Use the fetch_papers tool when you need additional paper content
- Provide accurate, detailed responses based on the actual paper content
- When referencing papers, use their actual titles from the filenames
- Prioritize full paper content over abstracts when available"""
    
    messages = [{"role": "system", "content": system_prompt}]
    
    # Add conversation history
    for user_msg, assistant_msg in history:
        if user_msg and user_msg.strip():
            messages.append({"role": "user", "content": user_msg.strip()})
        if assistant_msg and assistant_msg.strip():
            messages.append({"role": "assistant", "content": assistant_msg.strip()})
    
    # Add current user message
    messages.append({"role": "user", "content": message.strip()})
    
    # Truncate if needed
    messages = truncate_conversation_history(messages)
    
    try:
        model = AVAILABLE_MODELS.get(model_name, "gpt-4o-mini")
        
        # Initial response with tool support
        response = client.chat.completions.create(
            model=model,
            messages=messages,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            tools=[FETCH_PAPERS_TOOL],
            tool_choice="auto",
            stream=True
        )
        
        # Collect the response and handle tool calls
        full_response = ""
        tool_calls = []
        current_tool_call = None
        
        for chunk in response:
            if hasattr(chunk.choices[0], 'delta'):
                delta = chunk.choices[0].delta
                
                # Handle regular content
                if delta.content is not None:
                    full_response += delta.content
                    yield full_response
                
                # Handle tool calls
                if delta.tool_calls:
                    for tool_call_chunk in delta.tool_calls:
                        if tool_call_chunk.id:
                            # New tool call
                            if current_tool_call:
                                tool_calls.append(current_tool_call)
                            current_tool_call = {
                                "id": tool_call_chunk.id,
                                "type": "function",
                                "function": {
                                    "name": tool_call_chunk.function.name if tool_call_chunk.function else "",
                                    "arguments": ""
                                }
                            }
                        
                        if current_tool_call and tool_call_chunk.function:
                            if tool_call_chunk.function.arguments:
                                current_tool_call["function"]["arguments"] += tool_call_chunk.function.arguments
        
        # Add final tool call if exists
        if current_tool_call:
            tool_calls.append(current_tool_call)
        
        # Process tool calls if any
        if tool_calls:
            # Add the assistant's message with tool calls
            messages.append({
                "role": "assistant",
                "content": full_response if full_response else None,
                "tool_calls": tool_calls
            })
            
            # Execute tool calls
            for tool_call in tool_calls:
                function_name = tool_call["function"]["name"]
                
                if function_name == "fetch_papers":
                    try:
                        # Parse arguments
                        arguments = json.loads(tool_call["function"]["arguments"])
                        filenames = arguments.get("filenames", [])
                        
                        # Fetch papers
                        papers_content = fetch_papers(filenames)
                        
                        # Add tool response to messages
                        tool_response = {
                            "role": "tool",
                            "tool_call_id": tool_call["id"],
                            "content": json.dumps(papers_content)
                        }
                        messages.append(tool_response)
                        
                    except Exception as e:
                        tool_response = {
                            "role": "tool",
                            "tool_call_id": tool_call["id"],
                            "content": f"Error: {str(e)}"
                        }
                        messages.append(tool_response)
            
            # Get final response with tool results
            final_response = client.chat.completions.create(
                model=model,
                messages=messages,
                max_tokens=max_tokens,
                temperature=temperature,
                top_p=top_p,
                stream=True
            )
            
            # Stream the final response
            final_text = ""
            for chunk in final_response:
                if hasattr(chunk.choices[0], 'delta') and chunk.choices[0].delta.content is not None:
                    final_text += chunk.choices[0].delta.content
                    yield full_response + "\n\n" + final_text if full_response else final_text
                
    except Exception as e:
        error_message = f"Error: {str(e)}"
        if "api_key" in str(e).lower():
            error_message = "Error: Invalid or missing OpenAI API key."
        elif "quota" in str(e).lower():
            error_message = "Error: API quota exceeded."
        elif "rate" in str(e).lower():
            error_message = "Error: Rate limit exceeded."
        yield error_message

def chat_fn(message, history, model_name, max_tokens, temperature, top_p):
    """Handle the entire chat interaction."""
    if not message.strip():
        return history
    
    history.append([message, ""])
    
    for response in respond(message, history[:-1], model_name, max_tokens, temperature, top_p):
        history[-1][1] = response
        yield history

def clear_history() -> tuple:
    """Clear the conversation history."""
    return [], ""

        # Create the Gradio interface
with gr.Blocks(
    title="📚 AI Research Paper Chatbot",
    theme=gr.themes.Soft(),
    css="""
    .gradio-container {
        max-width: 1200px !important;
        margin: auto !important;
    }
    """
) as demo:
    gr.Markdown(
        """
        # 📚 AI Research Paper Chatbot
        
        Chat with an AI assistant that can intelligently retrieve and analyze research papers.
        """
    )
    
    with gr.Row():
        with gr.Column(scale=3):
            chatbot = gr.Chatbot(
                height=500,
                show_label=False,
                container=True,
                bubble_full_width=False
            )
            
            with gr.Row():
                msg = gr.Textbox(
                    placeholder="Type your message here...",
                    show_label=False,
                    container=False,
                    scale=9
                )
                submit_btn = gr.Button("Send", variant="primary", scale=1)
                clear_btn = gr.Button("Clear", variant="secondary", scale=1)
        
        with gr.Column(scale=1):
            gr.Markdown("### ⚙️ Settings")
            
            model_dropdown = gr.Dropdown(
                choices=list(AVAILABLE_MODELS.keys()),
                value="GPT-4o",
                label="Model",
                info="Select the AI model to use"
            )
            
            max_tokens_slider = gr.Slider(
                minimum=1,
                maximum=4096,
                value=1024,
                step=1,
                label="Max Tokens",
                info="Maximum response length"
            )
            
            temperature_slider = gr.Slider(
                minimum=0.0,
                maximum=2.0,
                value=0.7,
                step=0.1,
                label="Temperature",
                info="Creativity level"
            )
            
            top_p_slider = gr.Slider(
                minimum=0.0,
                maximum=1.0,
                value=1.0,
                step=0.05,
                label="Top-p",
                info="Response diversity"
            )
            
            gr.Markdown("### 💡 Examples")
            example_btn1 = gr.Button("What papers discuss AI's impact on employment?", size="sm")
            example_btn2 = gr.Button("Show me the full paper about AI companions", size="sm")
            example_btn3 = gr.Button("Compare findings on AI in education", size="sm")
    
    # Event handlers
    msg.submit(
        chat_fn,
        [msg, chatbot, model_dropdown, max_tokens_slider, temperature_slider, top_p_slider],
        [chatbot],
        show_progress=True
    ).then(
        lambda: "",
        outputs=[msg]
    )
    
    submit_btn.click(
        chat_fn,
        [msg, chatbot, model_dropdown, max_tokens_slider, temperature_slider, top_p_slider],
        [chatbot],
        show_progress=True
    ).then(
        lambda: "",
        outputs=[msg]
    )
    
    clear_btn.click(clear_history, outputs=[chatbot, msg])
    
    # Example handlers
    example_btn1.click(lambda: "What papers discuss AI's impact on employment?", outputs=msg)
    example_btn2.click(lambda: "Show me the full paper about AI companions", outputs=msg)
    example_btn3.click(lambda: "Compare findings on AI in education", outputs=msg)

if __name__ == "__main__":
    if not os.getenv("OPENAI_API_KEY"):
        print("⚠️  Warning: OPENAI_API_KEY environment variable not set!")
    
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,
        show_error=True,
        quiet=False
    )