app.py

import os

import gradio as gr
import spaces
from dotenv import load_dotenv
from PyPDF2 import PdfReader

from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_cohere import CohereEmbeddings, ChatCohere
from langchain_community.vectorstores import FAISS
from langchain_classic.chains import ConversationalRetrievalChain


load_dotenv()


# =========================================================
# 1. ZERO-GPU STARTUP CHECK
# =========================================================

@spaces.GPU(duration=1)
def zerogpu_check():
    return None


# =========================================================
# 2. EXTRACT TEXT FROM PDF FILES
# =========================================================

def get_pdf_text(pdf_files):
    text = ""

    for pdf_file in pdf_files:
        pdf_reader = PdfReader(pdf_file)

        for page in pdf_reader.pages:
            text += page.extract_text() or ""

    return text


# =========================================================
# 3. SPLIT TEXT INTO CHUNKS
# =========================================================

def get_text_chunks(text):
    text_splitter = RecursiveCharacterTextSplitter(
        separators=["\n\n", "\n", " ", ""],
        chunk_size=1000,
        chunk_overlap=200,
        length_function=len
    )

    return text_splitter.split_text(text)


# =========================================================
# 4. CREATE COHERE EMBEDDINGS
# =========================================================

def get_embeddings():
    return CohereEmbeddings(
        model="embed-english-v3.0",
        user_agent="langchain"
    )


# =========================================================
# 5. CREATE CONVERSATIONAL RETRIEVAL CHAIN
# =========================================================

def get_conversational_chain(vectorstore):
    llm = ChatCohere(
        model="command-r-08-2024",
        temperature=0
    )

    conversational_chain = ConversationalRetrievalChain.from_llm(
        llm=llm,
        retriever=vectorstore.as_retriever(),
        return_source_documents=False
    )

    return conversational_chain


# =========================================================
# 6. PROCESS UPLOADED DOCUMENTS
# =========================================================

def process_documents(pdf_files):

    if not pdf_files:
        return (
            None,
            [],
            "Please upload at least one PDF."
        )

    if not os.getenv("COHERE_API_KEY"):
        return (
            None,
            [],
            "Error: Cohere API key is missing."
        )

    try:
        raw_text = get_pdf_text(pdf_files)

        if not raw_text.strip():
            return (
                None,
                [],
                "No extractable text found in the PDFs."
            )

        chunks = get_text_chunks(raw_text)

        if not chunks:
            return (
                None,
                [],
                "No text chunks were created."
            )

        embeddings = get_embeddings()

        vectorstore = FAISS.from_texts(
            texts=chunks,
            embedding=embeddings
        )

        chain = get_conversational_chain(vectorstore)

        return (
            chain,
            [],
            "Documents processed successfully! You can now ask questions."
        )

    except Exception as e:
        return (
            None,
            [],
            f"Error processing documents: {str(e)}"
        )


# =========================================================
# 7. CONVERT GRADIO CHAT HISTORY TO LANGCHAIN FORMAT
# =========================================================

def convert_chat_history(history):
    chat_history = []
    pending_question = None

    for message in history or []:

        if isinstance(message, dict):
            role = message.get("role")
            content = message.get("content", "")

            if isinstance(content, list):
                content = " ".join(
                    str(item.get("text", ""))
                    if isinstance(item, dict)
                    else str(item)
                    for item in content
                )

            content = str(content)

            if role == "user":
                pending_question = content

            elif role == "assistant" and pending_question is not None:
                chat_history.append(
                    (pending_question, content)
                )
                pending_question = None

        elif isinstance(message, (list, tuple)) and len(message) == 2:
            user_message, assistant_message = message

            if user_message:
                chat_history.append(
                    (
                        str(user_message),
                        str(assistant_message or "")
                    )
                )

    return chat_history


# =========================================================
# 8. HANDLE USER QUESTIONS
# =========================================================

def handle_user_question(question, history, chain):

    history = list(history or [])

    if not question or not question.strip():
        return history, ""

    question = question.strip()

    if chain is None:
        history.append({
            "role": "user",
            "content": question
        })

        history.append({
            "role": "assistant",
            "content": "Please upload and process your PDFs first."
        })

        return history, ""

    try:
        chat_history = convert_chat_history(history)

        response = chain.invoke({
            "question": question,
            "chat_history": chat_history
        })

        answer = response.get(
            "answer",
            "No answer was generated."
        )

        if isinstance(answer, list):
            answer = " ".join(
                str(item.get("text", ""))
                if isinstance(item, dict)
                else str(item)
                for item in answer
            )
        else:
            answer = str(answer)

        history.append({
            "role": "user",
            "content": question
        })

        history.append({
            "role": "assistant",
            "content": answer
        })

        return history, ""

    except Exception as e:
        history.append({
            "role": "user",
            "content": question
        })

        history.append({
            "role": "assistant",
            "content": f"Error generating answer: {str(e)}"
        })

        return history, ""


# =========================================================
# 9. CLEAR CHAT
# =========================================================

def clear_chat():
    return []


# =========================================================
# 10. CUSTOM CSS
# =========================================================

custom_css = """
.gradio-container {
    background: #0e0f14 !important;
    color: #f8fafc !important;
    max-width: 100% !important;
    min-height: 100vh;
    padding: 16px !important;
}

#main-row {
    display: flex !important;
    align-items: flex-start !important;
    gap: 20px !important;
    max-width: 1500px;
    margin: 0 auto !important;
}

#sidebar {
    background: #24252f !important;
    border: 1px solid #343642 !important;
    border-radius: 14px !important;
    padding: 16px !important;
    height: fit-content !important;
    min-height: 0 !important;
    align-self: flex-start !important;
    margin-top: 40px;
}

#sidebar h2 {
    font-size: 1.15rem !important;
    margin: 0 0 12px 0 !important;
    color: #ffffff !important;
}

#sidebar [data-testid="file-upload"] {
    min-height: 110px !important;
}

#process-button {
    margin-top: 10px !important;
    border-radius: 8px !important;
    transition: transform 0.2s ease, box-shadow 0.2s ease;
}

#process-button:hover {
    transform: translateY(-1px);
    box-shadow: 0 4px 12px rgba(37, 99, 235, 0.3);
}

#status {
    margin-top: 10px !important;
}

#main-panel {
    padding: 0 8px !important;
    min-width: 0 !important;
}

#hero {
    background: linear-gradient(120deg, #0f172a 0%, #1d4ed8 100%);
    padding: 28px 32px;
    border-radius: 18px;
    box-shadow: 0 10px 30px rgba(29, 78, 216, 0.18);
    margin: 24px 0 18px 0;
    transition: box-shadow 0.2s ease;
}

#hero:hover {
    box-shadow: 0 12px 34px rgba(29, 78, 216, 0.28);
}

#hero h1 {
    color: white !important;
    font-size: clamp(1.5rem, 2.5vw, 2.3rem) !important;
    font-weight: 800 !important;
    line-height: 1.25 !important;
    margin: 0 !important;
}

#hero p {
    color: #dbeafe !important;
    font-size: 1rem !important;
    margin: 10px 0 0 0 !important;
}

#chatbot {
    border-radius: 12px !important;
    min-height: 280px !important;
    height: calc(100vh - 390px) !important;
    max-height: 650px !important;
    overflow: auto !important;
}

#question-box {
    margin-top: 12px !important;
}

#question-box textarea {
    min-height: 48px !important;
    border-radius: 10px !important;
}

#main-panel button {
    border-radius: 8px !important;
    transition: background 0.2s ease;
}

footer {
    display: none !important;
}

@media (max-width: 900px) {
    .gradio-container {
        padding: 12px !important;
    }

    #main-row {
        gap: 14px !important;
    }

    #sidebar {
        margin-top: 30px;
        padding: 12px !important;
    }

    #hero {
        padding: 22px;
    }

    #chatbot {
        height: 55vh !important;
    }
}

@media (max-width: 700px) {
    .gradio-container {
        padding: 8px !important;
    }

    #main-row {
        flex-direction: column !important;
        gap: 12px !important;
    }

    #sidebar {
        width: 100% !important;
        max-width: 100% !important;
        margin-top: 0 !important;
        padding: 14px !important;
    }

    #main-panel {
        width: 100% !important;
        max-width: 100% !important;
        padding: 0 !important;
    }

    #hero {
        margin: 8px 0 14px 0;
        padding: 20px;
        border-radius: 14px;
    }

    #hero h1 {
        font-size: 1.5rem !important;
    }

    #hero p {
        font-size: 0.9rem !important;
    }

    #chatbot {
        height: 55vh !important;
        min-height: 300px !important;
        max-height: none !important;
    }

    #question-box {
        margin-top: 10px !important;
    }
}
"""


# =========================================================
# 11. BUILD GRADIO INTERFACE
# =========================================================

with gr.Blocks(
    title="Chat with Documents"
) as demo:

    # State to store the conversational chain
    chain_state = gr.State(None)

    # Hidden event to satisfy ZeroGPU startup detection
    hidden_button = gr.Button(visible=False)
    hidden_button.click(
        fn=zerogpu_check,
        inputs=[],
        outputs=[]
    )

    with gr.Row(
        elem_id="main-row",
        equal_height=False
    ):

        # LEFT SIDEBAR
        with gr.Column(
            scale=1,
            min_width=250,
            elem_id="sidebar"
        ):

            gr.Markdown("## Upload Documents")

            pdf_upload = gr.File(
                file_types=[".pdf"],
                file_count="multiple",
                type="filepath",
                height=120
            )

            process_button = gr.Button(
                "Process Documents",
                variant="primary",
                elem_id="process-button"
            )

            status = gr.Textbox(
                label="Status",
                interactive=False,
                lines=2,
                elem_id="status"
            )

        # RIGHT CHAT AREA
        with gr.Column(
            scale=4,
            min_width=300,
            elem_id="main-panel"
        ):

            gr.HTML("""
                <div id="hero">
                    <h1>AI Chatbot for Your Documents</h1>
                    <p>
                        Upload a PDF and ask questions in natural language.
                    </p>
                </div>
            """)

            chatbot = gr.Chatbot(
                label="Chat with your PDFs",
                height=500,
                elem_id="chatbot"
            )

            question = gr.Textbox(
                label="Ask a question",
                placeholder="Ask a question about your PDFs",
                lines=1,
                max_lines=1,
                elem_id="question-box"
            )

            clear_button = gr.Button("Clear Chat")

    # Process PDF documents
    process_button.click(
        fn=process_documents,
        inputs=[pdf_upload],
        outputs=[chain_state, chatbot, status]
    )

    # Press Enter to send a question
    question.submit(
        fn=handle_user_question,
        inputs=[question, chatbot, chain_state],
        outputs=[chatbot, question]
    )

    # Clear conversation
    clear_button.click(
        fn=clear_chat,
        inputs=[],
        outputs=[chatbot]
    )


# =========================================================
# 12. LAUNCH APPLICATION
# =========================================================

if __name__ == "__main__":
    demo.queue().launch(
        theme=gr.themes.Base(),
        css=custom_css,
        server_name="0.0.0.0",
        server_port=7860,
        ssr_mode=False
    )