app.py

import os
import gradio as gr
from langchain.document_loaders import PyPDFLoader, YoutubeLoader, TextLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.chains import RetrievalQA
from langchain.chat_models import init_chat_model

# --- API KEY HANDLING ---
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") or os.getenv("openai")
if not OPENAI_API_KEY:
    raise ValueError("❌ OPENAI API Key not found. Please add it in Hugging Face secrets as 'OPENAI_API_KEY' or 'openai'.")

# --- PROCESSING FUNCTION ---
def process_inputs(pdf_file, youtube_url, txt_file, query):
    docs = []

    # Load PDF
    try:
        pdf_path = pdf_file.name
        pdf_loader = PyPDFLoader(pdf_path)
        docs.extend(pdf_loader.load())
    except Exception as e:
        return f"❌ Failed to load PDF: {e}"

    # Load YouTube Transcript (optional)
    yt_loaded = False
    if youtube_url:
        try:
            yt_loader = YoutubeLoader.from_youtube_url(youtube_url, add_video_info=False)
            docs.extend(yt_loader.load())
            yt_loaded = True
        except Exception as e:
            print(f"⚠️ YouTube transcript not loaded: {e}")

    # Load text transcript file (optional fallback)
    if not yt_loaded and txt_file is not None:
        try:
            txt_path = txt_file.name
            txt_loader = TextLoader(txt_path)
            docs.extend(txt_loader.load())
        except Exception as e:
            return f"❌ Failed to load transcript file: {e}"

    if not docs:
        return "❌ No documents could be loaded. Please check your inputs."

    # Split text into chunks
    splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=150)
    splits = splitter.split_documents(docs)

    # Embed documents
    embedding = OpenAIEmbeddings(model="text-embedding-3-large", api_key=OPENAI_API_KEY)
    db = FAISS.from_documents(splits, embedding)

    # Query using RetrievalQA
    llm = init_chat_model("gpt-4o-mini", model_provider="openai", api_key=OPENAI_API_KEY)
    qa = RetrievalQA.from_chain_type(llm, retriever=db.as_retriever())

    try:
        result = qa.invoke({"query": query})
        return result["result"]
    except Exception as e:
        return f"❌ Retrieval failed: {e}"

# --- GRADIO UI ---
with gr.Blocks() as demo:
    gr.Markdown("## 📚 Ask Questions from PDF + YouTube Transcript or .txt Upload")

    with gr.Row():
        pdf_input = gr.File(label="Upload PDF", file_types=[".pdf"])
        yt_input = gr.Textbox(label="YouTube URL (Optional)", placeholder="https://www.youtube.com/watch?v=...")
        txt_input = gr.File(label="Upload Transcript .txt (Optional fallback)", file_types=[".txt"])

    query_input = gr.Textbox(label="Your Question", placeholder="e.g., What did the document say about X?")
    output = gr.Textbox(label="Answer")

    run_button = gr.Button("Get Answer")
    run_button.click(fn=process_inputs, inputs=[pdf_input, yt_input, txt_input, query_input], outputs=output)

if __name__ == "__main__":
    demo.launch()