server/main.py

from fastapi import FastAPI, HTTPException, APIRouter
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import FileResponse
from fastapi.staticfiles import StaticFiles
from pydantic import BaseModel
from typing import List, Dict, Any, Optional
import os
import json
from workflow import create_workflow, run_workflow
import logging
from dotenv import load_dotenv
from langchain_openai import ChatOpenAI
from langchain.prompts import ChatPromptTemplate
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Qdrant
from langchain_openai.embeddings import OpenAIEmbeddings
from langchain_openai.chat_models import ChatOpenAI
from operator import itemgetter
from langchain.schema.output_parser import StrOutputParser
from langchain.schema.runnable import RunnablePassthrough


# Load environment variables
load_dotenv()

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Initialize components
openai_api_key = os.getenv("OPENAI_API_KEY")
if not openai_api_key:
    raise ValueError("OpenAI API key not configured")

# Initialize OpenAI components
chat_model = ChatOpenAI(
    model_name="gpt-3.5-turbo",
    temperature=0.7,
    openai_api_key=openai_api_key
)

# Define Pydantic models
class ChatMessage(BaseModel):
    content: str
    context: Optional[Dict[str, Any]] = None
    agent_type: Optional[str] = "believer"

class WorkflowResponse(BaseModel):
    debate_history: List[Dict[str, str]]
    supervisor_notes: List[str]
    supervisor_chunks: List[Dict[str, List[str]]]
    extractor_data: Dict[str, Any]
    final_podcast: Dict[str, Any]

class PodcastChatRequest(BaseModel):
    message: str

class PodcastChatResponse(BaseModel):
    response: str

# Initialize FastAPI app
app = FastAPI()

# Create API router
api_router = APIRouter(prefix="/api")

# Configure CORS
app.add_middleware(
    CORSMiddleware,
    allow_origins=["http://localhost:5173", "http://localhost:3000", "https://*.hf.space", "*"],
    allow_credentials=True,
    allow_methods=["GET", "POST", "PUT", "DELETE", "OPTIONS", "HEAD"],
    allow_headers=["*"],
    expose_headers=["Content-Type", "Content-Length"],
    max_age=600,
)

# Configure storage directories
audio_dir = os.path.join(os.path.dirname(__file__), "audio_storage")
os.makedirs(audio_dir, exist_ok=True)

context_dir = os.path.join(os.path.dirname(__file__), "context_storage")
os.makedirs(context_dir, exist_ok=True)

# Add transcripts directory
transcripts_dir = os.path.join(os.path.dirname(__file__), "transcripts")
os.makedirs(transcripts_dir, exist_ok=True)

# Initialize empty transcripts file if it doesn't exist
transcripts_file = os.path.join(transcripts_dir, "podcasts.json")
if not os.path.exists(transcripts_file):
    with open(transcripts_file, 'w') as f:
        json.dump([], f)

# API Routes
@api_router.post("/chat")
async def chat(message: ChatMessage):
    """Process a chat message."""
    try:
        # Get API key
        tavily_api_key = os.getenv("TAVILY_API_KEY")
        if not tavily_api_key:
            logger.error("Tavily API key not found")
            raise HTTPException(status_code=500, detail="Tavily API key not configured")

        # Initialize the workflow
        try:
            workflow = create_workflow(tavily_api_key)
            logger.info("Workflow created successfully")
        except Exception as e:
            logger.error(f"Error creating workflow: {str(e)}")
            raise HTTPException(status_code=500, detail=f"Error creating workflow: {str(e)}")
        
        # Run the workflow with context
        try:
            result = await run_workflow(
                workflow,
                message.content,
                agent_type=message.agent_type,
                context=message.context
            )
            logger.info("Workflow completed successfully")
            return result
        except Exception as e:
            logger.error(f"Error running workflow: {str(e)}")
            raise HTTPException(status_code=500, detail=f"Error running workflow: {str(e)}")
            
    except Exception as e:
        logger.error(f"Error in chat endpoint: {str(e)}", exc_info=True)
        raise HTTPException(status_code=500, detail=str(e))

@api_router.get("/audio-list")
async def list_audio_files():
    """List all available audio files."""
    try:
        files = os.listdir(audio_dir)
        audio_files = []
        for file in files:
            if file.endswith(('.mp3', '.wav')):
                file_path = os.path.join(audio_dir, file)
                audio_files.append({
                    "filename": file,
                    "path": f"/audio-files/{file}",
                    "size": os.path.getsize(file_path)
                })
        return audio_files if audio_files else []
    except Exception as e:
        logger.error(f"Error listing audio files: {str(e)}")
        return []

@api_router.get("/audio/{filename}")
async def get_audio_file(filename: str):
    """Get an audio file by filename."""
    try:
        file_path = os.path.join(audio_dir, filename)
        if not os.path.exists(file_path):
            logger.error(f"Audio file not found: {filename}")
            raise HTTPException(status_code=404, detail="File not found")
        return FileResponse(file_path, media_type="audio/mpeg")
    except Exception as e:
        logger.error(f"Error serving audio file: {str(e)}")
        raise HTTPException(status_code=500, detail=str(e))

@api_router.delete("/audio/{filename}")
async def delete_audio_file(filename: str):
    """Delete an audio file and its corresponding transcript."""
    try:
        # Check if file exists before attempting deletion
        file_path = os.path.join(audio_dir, filename)
        if not os.path.exists(file_path):
            logger.error(f"File not found for deletion: {filename}")
            raise HTTPException(status_code=404, detail="File not found")
        
        try:
            # Delete the audio file first
            os.remove(file_path)
            logger.info(f"Deleted audio file: {filename}")
            
            # Get all remaining audio files
            audio_files = [f for f in os.listdir(audio_dir) if f.endswith(('.mp3', '.wav'))]
            
            # Try to update transcripts if they exist
            transcripts_file = os.path.join(os.path.dirname(__file__), "transcripts", "podcasts.json")
            if os.path.exists(transcripts_file):
                with open(transcripts_file, 'r') as f:
                    transcripts = json.load(f)
                
                # Find the index of the deleted file in the original list
                try:
                    podcast_id = audio_files.index(filename) + 1
                    if len(transcripts) >= podcast_id:
                        transcripts.pop(podcast_id - 1)
                        with open(transcripts_file, 'w') as f:
                            json.dump(transcripts, f, indent=2)
                        logger.info(f"Updated transcripts after deletion")
                except ValueError:
                    logger.warning(f"Could not find podcast ID for {filename} in transcripts")
            
            return {"message": "File deleted successfully"}
            
        except Exception as e:
            logger.error(f"Error during file deletion process: {str(e)}")
            raise HTTPException(status_code=500, detail=str(e))
            
    except HTTPException as he:
        raise he
    except Exception as e:
        logger.error(f"Error in delete_audio_file: {str(e)}")
        raise HTTPException(status_code=500, detail=str(e))

@api_router.get("/podcast/{podcast_id}/context")
async def get_podcast_context(podcast_id: str):
    """Get or generate context for a podcast."""
    try:
        logger.info(f"Getting context for podcast {podcast_id}")
        context_path = os.path.join(context_dir, f"{podcast_id}_context.json")
        
        # If context exists, return it
        if os.path.exists(context_path):
            logger.info(f"Found existing context file at {context_path}")
            with open(context_path, 'r') as f:
                return json.load(f)
        
        # If no context exists, we need to create it from the podcast content
        logger.info("No existing context found, creating new context")
        
        # Get the audio files to find the podcast filename
        files = os.listdir(audio_dir)
        logger.info(f"Found {len(files)} files in audio directory")
        podcast_files = [f for f in files if f.endswith('.mp3')]
        logger.info(f"Found {len(podcast_files)} podcast files: {podcast_files}")
        
        if not podcast_files:
            logger.error("No podcast files found")
            raise HTTPException(status_code=404, detail="No podcast files found")
            
        # Find the podcast file that matches this ID
        try:
            podcast_index = int(podcast_id) - 1  # Convert 1-based ID to 0-based index
            if podcast_index < 0 or podcast_index >= len(podcast_files):
                raise ValueError(f"Invalid podcast ID: {podcast_id}, total podcasts: {len(podcast_files)}")
            podcast_filename = podcast_files[podcast_index]
            logger.info(f"Selected podcast file: {podcast_filename}")
        except (ValueError, IndexError) as e:
            logger.error(f"Invalid podcast ID: {podcast_id}, Error: {str(e)}")
            raise HTTPException(status_code=404, detail=f"Invalid podcast ID: {podcast_id}")
        
        # Extract topic from filename
        try:
            topic = podcast_filename.split('-')[0].replace('_', ' ')
            logger.info(f"Extracted topic: {topic}")
        except Exception as e:
            logger.error(f"Error extracting topic from filename: {podcast_filename}, Error: {str(e)}")
            raise HTTPException(status_code=500, detail=f"Error extracting topic from filename: {str(e)}")

        # Initialize OpenAI chat model for content analysis
        try:
            chat_model = ChatOpenAI(
                model_name="gpt-3.5-turbo",
                temperature=0.3,
                openai_api_key=openai_api_key
            )
            logger.info("Successfully initialized ChatOpenAI")
        except Exception as e:
            logger.error(f"Error initializing ChatOpenAI: {str(e)}")
            raise HTTPException(status_code=500, detail=f"Error initializing chat model: {str(e)}")

        # Create prompt template for content analysis
        prompt = ChatPromptTemplate.from_messages([
            ("system", """You are an expert content analyzer. Your task is to:
            1. Analyze the given topic and create balanced, factual content chunks about it
            2. Generate two types of chunks:
               - Believer chunks: Positive aspects, opportunities, and solutions related to the topic
               - Skeptic chunks: Challenges, risks, and critical questions about the topic
            3. Each chunk should be self-contained and focused on a single point
            4. Keep chunks concise (2-3 sentences each)
            5. Ensure all content is factual and balanced
            
            Format your response as a JSON object with two arrays:
            {{
                "believer_chunks": ["chunk1", "chunk2", ...],
                "skeptic_chunks": ["chunk1", "chunk2", ...]
            }}"""),
            ("human", "Create balanced content chunks about this topic: {topic}")
        ])

        # Generate content chunks
        chain = prompt | chat_model
        
        try:
            logger.info(f"Generating content chunks for topic: {topic}")
            response = await chain.ainvoke({
                "topic": topic
            })
            logger.info("Successfully received response from OpenAI")
            
            # Parse the response content as JSON
            try:
                content_chunks = json.loads(response.content)
                logger.info(f"Successfully parsed response JSON with {len(content_chunks.get('believer_chunks', []))} believer chunks and {len(content_chunks.get('skeptic_chunks', []))} skeptic chunks")
            except json.JSONDecodeError as e:
                logger.error(f"Error parsing response JSON: {str(e)}, Response content: {response.content}")
                raise HTTPException(status_code=500, detail=f"Error parsing content chunks: {str(e)}")
            
            # Create the context object
            context = {
                "topic": topic,
                "believer_chunks": content_chunks.get("believer_chunks", []),
                "skeptic_chunks": content_chunks.get("skeptic_chunks", [])
            }
            
            # Save the context
            try:
                with open(context_path, 'w') as f:
                    json.dump(context, f)
                    logger.info(f"Saved new context to {context_path}")
            except Exception as e:
                logger.error(f"Error saving context file: {str(e)}")
                raise HTTPException(status_code=500, detail=f"Error saving context file: {str(e)}")
            
            return context
            
        except Exception as e:
            logger.error(f"Error generating content chunks: {str(e)}")
            raise HTTPException(
                status_code=500,
                detail=f"Error generating content chunks: {str(e)}"
            )
            
    except HTTPException:
        raise
    except Exception as e:
        logger.error(f"Error in get_podcast_context: {str(e)}", exc_info=True)
        raise HTTPException(status_code=500, detail=str(e))

@api_router.post("/podcast-chat/{podcast_id}")
async def podcast_chat(podcast_id: str, request: PodcastChatRequest):
    """Handle chat messages for a specific podcast."""
    try:
        logger.info(f"Processing chat message for podcast {podcast_id}")
        logger.info(f"User message: {request.message}")
        
        # Get list of audio files
        audio_files = [f for f in os.listdir(audio_dir) if f.endswith('.mp3')]
        logger.info(f"Found {len(audio_files)} audio files: {audio_files}")
        
        # Convert podcast_id to zero-based index and get the filename
        try:
            podcast_index = int(podcast_id) - 1
            if podcast_index < 0 or podcast_index >= len(audio_files):
                logger.error(f"Invalid podcast index: {podcast_index} (total files: {len(audio_files)})")
                raise ValueError(f"Invalid podcast ID: {podcast_id}")
            podcast_filename = audio_files[podcast_index]
            logger.info(f"Found podcast file: {podcast_filename}")
        except ValueError as e:
            logger.error(f"Error converting podcast ID: {str(e)}")
            raise HTTPException(status_code=404, detail=str(e))
        
        # Extract topic from filename
        topic = podcast_filename.split('-')[0].replace('_', ' ')
        logger.info(f"Extracted topic: {topic}")
        
        # Path to transcripts file
        transcripts_file = os.path.join(os.path.dirname(__file__), "transcripts", "podcasts.json")
        
        # Check if transcripts file exists
        if not os.path.exists(transcripts_file):
            logger.error("Transcripts file not found")
            raise HTTPException(status_code=404, detail="Transcripts file not found")
            
        # Read transcripts
        try:
            with open(transcripts_file, 'r') as f:
                transcripts = json.load(f)
                logger.info(f"Loaded {len(transcripts)} transcripts")
                logger.info(f"Available topics: {[t.get('topic', 'NO_TOPIC') for t in transcripts]}")
        except json.JSONDecodeError as e:
            logger.error(f"Error decoding transcripts file: {str(e)}")
            raise HTTPException(status_code=500, detail="Error reading transcripts file")
        
        # Find matching transcript by topic
        podcast_transcript = None
        for transcript in transcripts:
            transcript_topic = transcript.get("topic", "").lower().strip()
            if transcript_topic == topic.lower().strip():
                podcast_transcript = transcript.get("podcastScript")
                logger.info(f"Found matching transcript for topic: {topic}")
                break
        
        if not podcast_transcript:
            logger.error(f"No transcript found for topic: {topic}")
            logger.error(f"Available topics: {[t.get('topic', 'NO_TOPIC') for t in transcripts]}")
            raise HTTPException(status_code=404, detail=f"No transcript found for topic: {topic}")
        
        logger.info(f"Found transcript for topic: {topic}")
        logger.info(f"Full transcript length: {len(podcast_transcript)} characters")
        logger.debug(f"Transcript preview: {podcast_transcript[:200]}...")

        # Split text into chunks
        text_splitter = RecursiveCharacterTextSplitter(
            chunk_size=1000,
            chunk_overlap=100,
            length_function=len,
            separators=["\n\n", "\n", ". ", " ", ""]
        )
        
        # Use split_text for strings instead of split_documents
        try:
            logger.info("Starting text splitting process...")
            chunks = text_splitter.split_text(podcast_transcript)
            logger.info(f"Successfully split transcript into {len(chunks)} chunks")
            
            # Log some sample chunks
            logger.info("\nSample chunks:")
            for i, chunk in enumerate(chunks[:3]):  # Log first 3 chunks
                logger.info(f"\nChunk {i+1}:")
                logger.info("=" * 50)
                logger.info(chunk)
                logger.info("=" * 50)
            
            if len(chunks) > 3:
                logger.info(f"... and {len(chunks) - 3} more chunks")
                
        except Exception as e:
            logger.error(f"Error splitting text into chunks: {str(e)}")
            raise HTTPException(status_code=500, detail=f"Error splitting text: {str(e)}")
        
        if not chunks:
            logger.error("No content chunks found in transcript")
            raise HTTPException(status_code=404, detail="No content chunks found in transcript")
            
        # Validate chunk sizes
        chunk_sizes = [len(chunk) for chunk in chunks]
        logger.info(f"\nChunk size statistics:")
        logger.info(f"Min chunk size: {min(chunk_sizes)} characters")
        logger.info(f"Max chunk size: {max(chunk_sizes)} characters")
        logger.info(f"Average chunk size: {sum(chunk_sizes)/len(chunk_sizes):.2f} characters")
        
        # Initialize embedding model
        embedding_model = OpenAIEmbeddings(
            model="text-embedding-3-small",
            openai_api_key=openai_api_key
        )
        
        # Create a unique collection name for this podcast
        collection_name = f"podcast_{podcast_id}"
        
        # Initialize Qdrant with local storage
        vectorstore = Qdrant.from_texts(
            texts=chunks,
            embedding=embedding_model,
            location=":memory:",  # Use in-memory storage
            collection_name=collection_name
        )
        logger.info(f"Created vector store for podcast {podcast_id}")
        
        # Configure the retriever with search parameters
        qdrant_retriever = vectorstore.as_retriever(
            search_type="similarity",  # Use simple similarity search
            search_kwargs={
                "k": 8,  # Increased from 5 to 8 chunks
                "score_threshold": 0.05  # Lowered threshold further for more matches
            }
        )

        base_rag_prompt_template = """\
        You are a helpful podcast assistant. Answer the user's question based on the provided context from the podcast transcript.
        If the context contains relevant information, use it to answer the question.
        If you can't find relevant information in the context to answer the question, say "I don't have enough information to answer that question."
        Keep your responses concise and focused on the question.
        
        Important: Even if only part of the context is relevant to the question, use that part to provide a partial answer rather than saying there isn't enough information.

        Context:
        {context}

        Question:
        {question}

        Answer the question using the information from the context above. If you find ANY relevant information, use it to provide at least a partial answer. Only say "I don't have enough information" if there is absolutely nothing relevant in the context.
        """

        base_rag_prompt = ChatPromptTemplate.from_template(base_rag_prompt_template)
        base_llm = ChatOpenAI(
            model="gpt-3.5-turbo",
            temperature=0.7,
            openai_api_key=openai_api_key
        )

        # Create the RAG chain
        def format_docs(docs):
            formatted = "\n\n".join(doc.page_content for doc in docs)
            logger.info(f"Formatted {len(docs)} documents into context of length: {len(formatted)}")
            return formatted

        # Add logging for the retrieved documents and final prompt
        def get_context_and_log(input_dict):
            try:
                logger.info("\nAttempting to retrieve relevant documents...")
                # Log the query being used
                logger.info(f"Query: {input_dict['question']}")
                
                # Use the newer invoke method instead of get_relevant_documents
                retrieved_docs = qdrant_retriever.invoke(input_dict["question"])
                logger.info(f"Successfully retrieved {len(retrieved_docs)} documents")
                
                if not retrieved_docs:
                    logger.warning("No documents were retrieved!")
                    return {"context": "No relevant context found.", "question": input_dict["question"]}
                
                # Log each retrieved document with its content and similarity score
                total_content_length = 0
                for i, doc in enumerate(retrieved_docs):
                    logger.info(f"\nDocument {i+1}:")
                    logger.info("=" * 50)
                    logger.info(f"Content: {doc.page_content}")
                    logger.info(f"Content Length: {len(doc.page_content)} characters")
                    logger.info(f"Metadata: {doc.metadata}")
                    logger.info("=" * 50)
                    total_content_length += len(doc.page_content)
                
                context = format_docs(retrieved_docs)
                
                # Log the final formatted context and question
                logger.info("\nRetrieval Statistics:")
                logger.info(f"Total documents retrieved: {len(retrieved_docs)}")
                logger.info(f"Total content length: {total_content_length} characters")
                logger.info(f"Average document length: {total_content_length/len(retrieved_docs):.2f} characters")
                
                logger.info("\nFinal Context and Question:")
                logger.info("=" * 50)
                logger.info("Context:")
                logger.info(f"{context}")
                logger.info("-" * 50)
                logger.info(f"Question: {input_dict['question']}")
                logger.info("=" * 50)
                
                if not context.strip():
                    logger.error("Warning: Empty context retrieved!")
                    return {"context": "No relevant context found.", "question": input_dict["question"]}
                
                return {"context": context, "question": input_dict["question"]}
            except Exception as e:
                logger.error(f"Error in get_context_and_log: {str(e)}")
                logger.error("Stack trace:", exc_info=True)
                return {"context": "Error retrieving context.", "question": input_dict["question"]}

        # Create the chain
        chain = (
            RunnablePassthrough()
            | get_context_and_log
            | base_rag_prompt
            | base_llm
        )

        # Get response with enhanced logging
        try:
            logger.info("\nGenerating response...")
            response = chain.invoke({"question": request.message})
            logger.info("=" * 50)
            logger.info("Final Response:")
            logger.info(f"{response.content}")
            logger.info("=" * 50)
            
            return PodcastChatResponse(response=response.content)
        except Exception as e:
            logger.error(f"Error generating response: {str(e)}")
            raise HTTPException(status_code=500, detail=f"Error generating response: {str(e)}")
        
    except HTTPException:
        raise
    except Exception as e:
        logger.error(f"Error in podcast chat: {str(e)}", exc_info=True)
        raise HTTPException(status_code=500, detail=str(e))

# Include the API router
app.include_router(api_router)

# Mount static directories
app.mount("/audio-files", StaticFiles(directory=audio_dir), name="audio")
app.mount("/", StaticFiles(directory="static", html=True), name="frontend")

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)