app.py

import os
import asyncio
import hashlib
from dotenv import load_dotenv
import gradio as gr

# Load environment variables
load_dotenv()
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")

from langchain_pypdf import PyPDFLoader                          # Fix: was langchain_community
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_core.prompts import PromptTemplate                # Fix: replaces load_qa_chain
from langchain_core.output_parsers import StrOutputParser        # Fix: replaces load_qa_chain
from langchain_google_genai import ChatGoogleGenerativeAI

# Pre-initialize Embeddings Model to prevent slow reloads on every upload
print("LOG: Pre-initializing HuggingFace Embeddings model...")
EMBEDDINGS = HuggingFaceEmbeddings(
    model_name="sentence-transformers/all-MiniLM-L6-v2",
    model_kwargs={"device": "cpu"},
)
print("LOG: Embeddings pipeline ready.")


def build_vectorstore_gradio(uploaded_file):
    if uploaded_file is None:
        return (
            None,
            """
            <div style="background: rgba(255, 255, 255, 0.02); border: 1px dashed rgba(255, 255, 255, 0.1); border-radius: 8px; padding: 12px; text-align: center;">
                <p style="color: #8A8D9F; margin: 0; font-size: 0.9rem;">Waiting for PDF Upload...</p>
            </div>
            """,
            gr.update(interactive=False, placeholder="Upload a PDF on the left first to enable questioning..."),
            gr.update(visible=False),
        )

    file_path = uploaded_file.name
    file_name = os.path.basename(file_path)

    # Read bytes to compute MD5 hash for unique tracking/logs
    with open(file_path, "rb") as f:
        file_bytes = f.read()
    file_hash = hashlib.md5(file_bytes).hexdigest()

    print(f"LOG: Parsing PDF '{file_name}' (Hash: {file_hash})...")

    try:
        loader = PyPDFLoader(file_path)
        docs = loader.load()

        chunks = RecursiveCharacterTextSplitter(
            chunk_size=750, chunk_overlap=75
        ).split_documents(docs)

        print(f"LOG: Splitting complete. Total chunks created: {len(chunks)}")
        vectorstore = FAISS.from_documents(chunks, EMBEDDINGS)
        print("LOG: Vector store indexing finished.")

        status_msg = f"""
        <div style="background: rgba(16, 185, 129, 0.1); border: 1px solid #10b981; border-radius: 8px; padding: 12px; margin-top: 10px;">
            <p style="color: #10b981; margin: 0; font-weight: 600;">✓ Index Built Successfully!</p>
            <p style="color: #cbd5e1; margin: 4px 0 0 0; font-size: 0.85rem;">
                <b>Document:</b> {file_name} <br>
                <b>Size:</b> {os.path.getsize(file_path) / 1024:.1f} KB <br>
                <b>Total Segments:</b> {len(chunks)} Chunks
            </p>
        </div>
        """

        return (
            (chunks, vectorstore),
            status_msg,
            gr.update(interactive=True, placeholder="Ask a question about the document..."),
            gr.update(visible=True),
        )
    except Exception as e:
        err_msg = f"""
        <div style="background: rgba(239, 68, 68, 0.1); border: 1px solid #ef4444; border-radius: 8px; padding: 12px; margin-top: 10px;">
            <p style="color: #ef4444; margin: 0; font-weight: 600;">⚠️ Failed to process PDF</p>
            <p style="color: #cbd5e1; margin: 4px 0 0 0; font-size: 0.85rem;">{str(e)}</p>
        </div>
        """
        return None, err_msg, gr.update(interactive=False), gr.update(visible=False)


def answer_question_gradio(question, state, api_key_input):
    api_key = api_key_input.strip() or GOOGLE_API_KEY
    if not api_key:
        return (
            "⚠️ **Error:** `GOOGLE_API_KEY` is missing. Please add it to your `.env` file or enter it in the input field.",
            "",
        )

    if not state:
        return "⚠️ **Error:** No active document. Please upload a PDF file first.", ""

    if not question.strip():
        return "⚠️ Please type in a question.", ""

    chunks, vectorstore = state

    try:
        # Retrieve context
        relevant_docs = vectorstore.similarity_search(question, k=4)

        # Load LLM
        llm = ChatGoogleGenerativeAI(
            google_api_key=api_key,
            model="gemini-2.5-flash",
            temperature=0.3,
            max_retries=0,
        )

        # Modern LCEL chain — replaces deprecated load_qa_chain
        prompt = PromptTemplate.from_template(
            "Use the following context to answer the question as accurately as possible.\n\n"
            "Context:\n{context}\n\n"
            "Question: {question}\n\n"
            "Answer:"
        )
        chain = prompt | llm | StrOutputParser()
        context = "\n\n".join(doc.page_content for doc in relevant_docs)
        answer = chain.invoke({"context": context, "question": question})

        # Format sources as HTML/Markdown citations
        sources_html = ""
        for i, doc in enumerate(relevant_docs, 1):
            page = doc.metadata.get("page", 0) + 1
            sources_html += f"""
            <div style="background: rgba(255, 255, 255, 0.02); border-left: 3px solid #8b5cf6; padding: 10px; margin-bottom: 12px; border-radius: 6px; border: 1px solid rgba(255, 255, 255, 0.04);">
                <span style="font-weight: 600; color: #a78bfa; font-size: 0.82rem; background: rgba(139, 92, 246, 0.1); padding: 2px 6px; border-radius: 4px;">Chunk {i} — Page {page}</span>
                <p style="margin: 6px 0 0 0; font-size: 0.88rem; color: #cbd5e1; font-style: italic; line-height: 1.4;">
                    "...{doc.page_content.strip()}..."
                </p>
            </div>
            """

        return answer, sources_html

    except Exception as e:
        err = str(e)
        if "429" in err or "quota" in err.lower():
            return (
                "⚠️ **Rate Limit Hit:** You have exceeded the free tier quota for Gemini API. Please wait a minute and try again.",
                "",
            )
        return f"⚠️ **Error:** {err}", ""


# Custom Premium CSS Styling (DocuMind Theme)
custom_css = """
body {
    background-color: #0d0e15 !important;
    color: #f3f4f6 !important;
}
.gradio-container {
    max-width: 1200px !important;
    margin: 0 auto !important;
    font-family: 'Plus Jakarta Sans', -apple-system, sans-serif !important;
}
/* Title and Header styling */
.header-container {
    text-align: center;
    padding: 2rem 0;
    margin-bottom: 1.5rem;
    border-bottom: 1px solid rgba(255, 255, 255, 0.05);
}
.header-container h1 {
    font-size: 3rem;
    font-weight: 800;
    background: linear-gradient(135deg, #3B82F6 0%, #8B5CF6 50%, #06B6D4 100%);
    -webkit-background-clip: text;
    -webkit-text-fill-color: transparent;
    letter-spacing: -1.5px;
    margin: 0;
}
.header-container p {
    color: #8A8D9F;
    font-size: 1.15rem;
    margin: 8px 0 0 0;
}
/* Primary Button Gradient */
button.primary {
    background: linear-gradient(135deg, #8B5CF6 0%, #3B82F6 100%) !important;
    color: white !important;
    border: none !important;
    font-weight: 600 !important;
    transition: all 0.2s ease !important;
    border-radius: 8px !important;
}
button.primary:hover {
    transform: translateY(-1px);
    box-shadow: 0 4px 12px rgba(139, 92, 246, 0.35) !important;
}
"""

with gr.Blocks(title="DocuMind AI") as demo:
    gr.HTML(
        """
        <div class="header-container">
            <h1>DocuMind AI</h1>
            <p>Interactive PDF Question Answering Powered by LangChain, FAISS & Gemini</p>
        </div>
        """
    )

    # State to hold chunks and vectorstore
    doc_state = gr.State(None)

    with gr.Row():
        # Sidebar / Left column
        with gr.Column(scale=1, min_width=320):
            gr.Markdown("### 📂 Upload Document")
            pdf_file = gr.File(
                label="Select or Drop PDF File",
                file_types=[".pdf"],
                file_count="single",
            )

            status_panel = gr.HTML(
                """
                <div style="background: rgba(255, 255, 255, 0.02); border: 1px dashed rgba(255, 255, 255, 0.1); border-radius: 8px; padding: 12px; text-align: center;">
                    <p style="color: #8A8D9F; margin: 0; font-size: 0.9rem;">Waiting for PDF Upload...</p>
                </div>
                """
            )

            gr.Markdown("### 🔑 API Authentication")
            api_key_box = gr.Textbox(
                label="Google API Key Override (Optional)",
                placeholder="AIzaSy... (Leave empty to use .env key)",
                type="password",
            )

            gr.Markdown(
                """
                <div style="font-size: 0.82rem; color: #8A8D9F; margin-top: 15px; border-top: 1px solid rgba(255, 255, 255, 0.05); padding-top: 10px; line-height: 1.5;">
                    💡 <b>Private & Local Processing:</b><br>
                    Your document is parsed locally inside the workspace environment. The FAISS vector database operates securely in memory.
                </div>
                """
            )

        # Workspace Panel / Right column
        with gr.Column(scale=2):
            gr.Markdown("### 💬 Ask Questions")
            query_box = gr.Textbox(
                label="Ask a question about the document:",
                placeholder="Upload a PDF on the left first to enable questioning...",
                interactive=False,
                lines=2,
            )

            with gr.Row():
                submit_btn = gr.Button("Submit Question", variant="primary")
                clear_btn = gr.Button("Clear Workspace")

            gr.Markdown("### 💡 Answer Output")
            answer_box = gr.Markdown(
                "*(Upload a PDF and submit your question to view the AI output here.)*"
            )

            sources_accordion = gr.Accordion(
                "📚 Relevant Source Citations Used", open=True, visible=False
            )
            with sources_accordion:
                sources_box = gr.HTML("No document sources loaded.")

    # Wiring Event Handlers
    pdf_file.change(
        fn=build_vectorstore_gradio,
        inputs=[pdf_file],
        outputs=[doc_state, status_panel, query_box, sources_accordion],
    )

    # When clear is clicked
    def clear_workspace():
        return (
            "",
            "*(Upload a PDF and submit your question to view the AI output here.)*",
            "No document sources loaded.",
        )

    clear_btn.click(
        fn=clear_workspace,
        inputs=[],
        outputs=[query_box, answer_box, sources_box],
    )

    # Submitting questions
    submit_btn.click(
        fn=answer_question_gradio,
        inputs=[query_box, doc_state, api_key_box],
        outputs=[answer_box, sources_box],
    )
    query_box.submit(
        fn=answer_question_gradio,
        inputs=[query_box, doc_state, api_key_box],
        outputs=[answer_box, sources_box],
    )

if __name__ == "__main__":
    asyncio.set_event_loop(asyncio.new_event_loop())

    demo.launch(
        server_name="127.0.0.1",
        server_port=7860,
        share=False,
        theme=gr.themes.Default(),
        css=custom_css,
    )