app.py

import gradio as gr

from langchain_community.document_loaders import PyPDFLoader
from langchain_community.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.chains import RetrievalQA
from langchain_openai import ChatOpenAI


def run_qa(pdf_path, question):
    if pdf_path is None or question.strip() == "":
        return "Please upload a PDF and enter a question."

    # Load PDF
    loader = PyPDFLoader(pdf_path)
    docs = loader.load()

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

    # Create embeddings + vector store
    embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
    vectordb = FAISS.from_documents(chunks, embeddings)

    # LLM
    llm = ChatOpenAI(temperature=0)

    # Retrieval QA chain
    qa = RetrievalQA.from_chain_type(
        llm=llm,
        retriever=vectordb.as_retriever(),
        return_source_documents=True
    )

    # Newer LangChain-safe call
    result = qa.invoke({"query": question})

    answer_text = result.get("result", "")
    source_docs = result.get("source_documents", [])

    sources = "\n\n".join([d.page_content[:500] for d in source_docs[:2]])

    return f"### Answer\n{answer_text}\n\n---\n### Sources\n{sources}"


with gr.Blocks(title="Agentic Document Intelligence") as demo:
    gr.Markdown("# 📄 Agentic Document Intelligence\nUpload a PDF and ask questions using RAG.")

    pdf = gr.File(label="Upload PDF", type="filepath")
    question = gr.Textbox(label="Ask a question")
    output = gr.Markdown()

    btn = gr.Button("Run")
    btn.click(run_qa, inputs=[pdf, question], outputs=output)

demo.launch()