app.py

# Standard Library Imports
import os
import uuid
import time

# Third-Party Libraries
import requests
import pandas as pd
from dotenv import load_dotenv
from tenacity import retry, stop_after_delay, wait_fixed, RetryError
import gradio as gr

# LangChain Imports
from langchain.chat_models import ChatOpenAI
from langchain.chains import ConversationalRetrievalChain
from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import Chroma
from langchain.text_splitter import SpacyTextSplitter
from langchain.document_loaders import TextLoader
from langchain.memory import ConversationBufferMemory
from langchain.agents import initialize_agent, Tool
from langchain.agents.agent_types import AgentType

class AbbyyVantage:
    """
    A client to interact with the ABBYY Vantage public API.
    Handles authentication, skill listing, transaction initiation, and result retrieval.
    """

    def __init__(self, client_id, client_secret, region="au"):
        """
        Initializes the AbbyyVantageClient by authenticating using client credentials.

        Args:
            client_id (str): Your ABBYY Vantage client ID.
            client_secret (str): Your ABBYY Vantage client secret.
            region (str): ABBYY Vantage region ('eu', 'us', 'au', etc.). Defaults to 'au'.

        Raises:
            Exception: If authentication fails or access token is not returned.
        """
        self.client_id = client_id
        self.client_secret = client_secret
        self.token_url = f"https://vantage-{region}.abbyy.com/auth2/connect/token"
        self.api_base = f"https://vantage-{region}.abbyy.com/api/publicapi/v1"

        try:
            # Prepare data for token request using client credentials
            data = {
                'grant_type': 'client_credentials',
                'client_id': self.client_id,
                'client_secret': self.client_secret
            }

            # Request access token from ABBYY OAuth2 endpoint
            res = requests.post(self.token_url, data=data)
            res.raise_for_status()

            # Extract access token from response
            token = res.json().get('access_token')
            if not token:
                raise ValueError("No access token returned from ABBYY")

            # Set authorization headers for future API calls
            self._headers = {
                "Authorization": f"Bearer {token}",
                "accept": "application/json"
            }
        except Exception as e:
            print(f"Error during authentication: {e}")
            raise

    def get_skills(self):
        """
        Retrieves a list of available document processing skills from ABBYY Vantage.

        Returns:
            dict or None: A JSON object containing skill metadata or None if the request fails.
        """
        try:
            # Send GET request to fetch all available skills
            res = requests.get(f'{self.api_base}/skills', headers=self._headers)
            res.raise_for_status()
            return res.json()
        except Exception as e:
            print(f"Failed to fetch skills: {e}")
            return None

    def process_document(self, file_path, skill_id):
        """
        Starts a new transaction by uploading a file to be processed using a specific skill.

        Args:
            file_path (str): Path to the local PDF file to be uploaded.
            skill_id (str): The ID of the skill to be used for processing.

        Returns:
            str or None: The transaction ID returned by the API or None if the request fails.
        """
        try:
            # Prepare API URL with query parameter for the skill ID
            url = f"{self.api_base}/transactions/launch?skillId={skill_id}"

            # Open the file in binary mode for upload
            with open(file_path, "rb") as f:
                files = {
                    "Files": (os.path.basename(file_path), f, "application/pdf")
                }

                # Post the file to ABBYY API to start a transaction
                res = requests.post(url, headers=self._headers, files=files)
                res.raise_for_status()

                # Extract and return the transaction ID
                return res.json().get('transactionId')
        except Exception as e:
            print(f"Failed to start transaction: {e}")
            return None

    def get_document_results(self, transaction_id, output_path="result_file.txt"):
        """
        Checks the transaction status and downloads the result file if processing is complete.

        Args:
            transaction_id (str): The transaction ID to monitor.
            output_path (str): Local file path to save the result file. Defaults to "result_file.txt".

        Returns:
            str or None: Path to the saved result file, or None if processing is incomplete or fails.
        """
        try:
            # Get transaction status and metadata
            url = f"{self.api_base}/transactions/{transaction_id}"
            res = requests.get(url, headers=self._headers)
            res.raise_for_status()
            data = res.json()
        except Exception as e:
            print(f"Failed to fetch transaction details: {e}")
            return None

        # Extract processing status
        status = data.get('status')
        print(f"Transaction status: {status}")

        # Handle status outcomes
        if status == 'Processing':
            print("File is still being processed. Try again later.")
            return 'Processing'
        elif status != 'Processed':
            print(f"Unexpected status: {status}")
            return f"Unexpected status: {status}"

        try:
            # Navigate to the result file ID in the JSON structure
            file_id = data['documents'][0]['resultFiles'][0]['fileId']

            # Build the download URL using transaction ID and file ID
            download_url = f"{self.api_base}/transactions/{transaction_id}/files/{file_id}/download"

            # Download the result file
            res = requests.get(download_url, headers=self._headers)
            res.raise_for_status()

            # Save the file to the specified path
            with open(output_path, 'wb') as f:
                f.write(res.content)

            print(f"File downloaded and saved to: {output_path}")
            return 'Processed'

        except (KeyError, IndexError) as e:
            print(f"Error accessing file ID in response JSON: {e}")
        except Exception as e:
            print(f"Failed to download or save file: {e}")

# df = pd.DataFrame(client.get_skills())
# df

# ----------- Process OCR & Setup Retrieval Agent -------------
def process_pdf_ocr(file):
    print('process_pdf_ocr', file)
    client = AbbyyVantage(client_id=os.getenv("ABBY_CLIENT_ID"),
                          client_secret=os.getenv("ABBY_CLIENT_SECRET"),
                          region="au"  # or "us", "au", etc.
                          )
    skill_id = '1681402d-2931-41cb-9717-bb7612bc09aa'
    trans_id = client.process_document(file_path=file, skill_id=skill_id)

    @retry(stop=stop_after_delay(60), wait=wait_fixed(3))
    def wait_for_processing():
        status = client.get_document_results(trans_id, output_path="/tmp/result_file.txt")
        print(f"Status: {status}")
        if status == 'Processed':
            print("|-- Processed")
            return status
        raise Exception("Still Processing")

    try:
        status = wait_for_processing()
        print("|--OCR Successful")
        setup_agent("/tmp/result_file.txt")
        print("|--Chatbot is ready")
        return "OCR Successful. Chatbot is ready"
    except RetryError:
        print("|--OCR Failed or Timed Out")
        return "OCR Failed or Timed Out"


# Global state
retrieval_chain = None
agent_executor = None


# ----------- Setup LangChain Retrieval Agent -------------
def setup_agent(file):
    global retrieval_chain, agent_executor

    if not os.path.exists(file):
        return "Please process a PDF first."

    loader = TextLoader(file)
    documents = loader.load()

    splitter = SpacyTextSplitter()
    chunks = splitter.split_documents(documents)

    embeddings = OpenAIEmbeddings()
    vectordb = Chroma.from_documents(chunks, embedding=embeddings, collection_name=f"temp_collection_{uuid.uuid4().hex}")
    retriever = vectordb.as_retriever(search_kwargs={"k": 10})

    memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)

    retrieval_chain = ConversationalRetrievalChain.from_llm(
        llm=ChatOpenAI(model="gpt-3.5-turbo"),
        retriever=retriever,
        memory=memory,
        return_source_documents=False
    )

    tools = [
            Tool(
                name="PolicyRetrievalRAG",
                func=retrieval_chain.run,
                description="Use this tool to retrieve the most relevant clauses from the insurance policy based on the user's question."
            )
    ]

    agent_executor = initialize_agent(
        tools=tools,
        llm=ChatOpenAI(model="gpt-4o"),
        agent=AgentType.CHAT_CONVERSATIONAL_REACT_DESCRIPTION,
        memory=memory,
        handle_parsing_errors=True, 
        verbose=True
    )

    print("|--Agent setup complete")
    return "Agent is ready."

# ----------- Chat Interface Handler -------------
def ask_question(message, history):
    if agent_executor is None:
        return "❗ Chatbot not ready. Please upload and process a PDF first."

    advisory_prompt = (
        "Use the PolicyRetrievalRAG tool to extract the most relevant clauses from the policy document. "
        "Always base your response strictly on the actual policy content — do not fabricate or assume.\n\n"
    
        "Respond strictly in this structured format using **Markdown** with **emojis**, and wrap everything inside a `Final Answer:` block for compatibility:\n\n"
    
        "Final Answer:\n"
        "### 📄 Policy Details\n"
        "- Quote or paraphrase the most relevant clauses from the policy.\n\n"
    
        "### 🧠 Simplified Explanation\n"
        "- Translate the policy language into simple, user-friendly language.\n\n"
    
        "### 💡 Advisor’s Practical Tip\n"
        "- Give actionable tips to help the user get the most from their policy.\n\n"
    
        "### ⚠️ Caveats and Exclusions\n"
        "- Mention any exclusions, limitations, or waiting periods.\n\n"
    
        "Your tone should be empathetic and clear — like a smart, helpful insurance advisor.\n"
        "Always include **all four** sections.\n\n"
    
        "---\n"
        "### ✅ Example Response:\n"
        "Final Answer:\n"
        "### 📄 Policy Details\n"
        "- The policy includes a 24-month waiting period for pre-existing conditions.\n\n"
    
        "### 🧠 Simplified Explanation\n"
        "- You need to wait 2 years before any claims related to older health issues are covered.\n\n"
    
        "### 💡 Advisor’s Practical Tip\n"
        "- Explore top-up plans that might waive or reduce the waiting period.\n\n"
    
        "### ⚠️ Caveats and Exclusions\n"
        "- Conditions like diabetes and hypertension are counted as pre-existing, so they'll be excluded during the waiting period.\n"
    )


    
    prompt = f"{advisory_prompt}\nQuestion: {message}"

    try:
        response = agent_executor.run(prompt)
        return response
    except Exception as e:
        return f"❌ Error: {str(e)}"

# ----------- Gradio UI -------------
with gr.Blocks(theme='shivi/calm_seafoam', title="📄 Insurance Policy AIdvisor") as demo:
    gr.Markdown("# Insurance Policy AIdvisor App")
    gr.Markdown("### Upload policy and converse")
    with gr.Tab("📄 Upload PDF"):
        gr.Markdown("### Upload a PDF. And Intellignet Automation Processng will automatically processing it using ABBYY Vantage, ChromaDB and LangChain")
        pdf_file = gr.File(label="📤 Upload a PDF", file_types=[".pdf"])
        ocr_status = gr.Textbox(label="Processing Status", interactive=False)

        pdf_file.change(process_pdf_ocr, inputs=[pdf_file], outputs=[ocr_status])

        gr.Examples(
            examples=[["small-insudoc.pdf"],["Principal-Sample-Life-Insurance-Policy.pdf"], ["NIC - National Mediclaim Plus Policy.pdf"]],
            inputs=[pdf_file],
            label="Example PDFs"
        )

    with gr.Tab("💬 Chatbot"):
        chat = gr.ChatInterface(fn=ask_question,
                                title = "🤖 AIdvisor",
                                chatbot=gr.Chatbot(
                                        avatar_images=(
                                                        "https://em-content.zobj.net/source/twitter/141/parrot_1f99c.png",  # User
                                                        "https://em-content.zobj.net/source/twitter/141/robot-face_1f916.png"  # Bot
                                        )
                                )
               )
        
        gr.Examples(
            examples=[
                "In what forms are the certificate avalaible?",
                "How many employees should enroll if the member is to not contribute premium?",
                "Can insurer contest this policy?",
                "when can insurer make changes to the policy?",
                "I gave incorrect age in the policy, what to do now?",
                "Can the data I filled in the application form to get the insurance policy be used against me?"
            ],
            inputs=chat.textbox
        )
        
    with gr.Tab("System Design"):
        gr.Image(value="AIdvisor_devcon.png")
        
    with gr.Tab("UML-System Diagram"):
        gr.Markdown("[![](https://mermaid.ink/img/pako:eNrtVl1v2jAU_SuWn1qJIkIIhUirRGEgHrZVDCZ1QkImMcFqYme2s41V_e-7jhOapKnW91VIwbHP_b7HuY84ECHFPlb0R0Z5QGeMRJIkW45QSqRmAUsJ12i5mk3WiKh8sWyebhSV5tD8N88WkoRMbJbmvFw3MakUAVVql4aHnQhyVcvZXRM12e9Pp2-wIhE1kMnt7f09KjaaYEV1lu7gANYGmy_mJNBCnprY6VGKhMxuDbBcNzErqiWjP22cq8misrEWIn7ha83wlptzkmnBs2RvcyRpoJGM9hd9x-mg50ev27805wh9FpqimB40Eoc8tb7NPprH4peF2LJc3dw0U-ijTRoLEiISx-huNlcW30y0kaym1UeTTB_BYxYQME54iCKqkRYPlL9VQ4kIRZAloOriwGLaQeqBxfGOhUVwtWJetUagJeEKKsYEBzkrFguRog34B1FZPA2RgBqwhIpMW9BbvISwzh7uJFVZrNVF3eJlqe0NvipNdKbQh9IrxiPj1tlHq4rysC36NoVle6HCt9ezX-l0H301LyCTC5MYEdt-RrbKiFaTC6j0rEjJa_ZeGGwINI2UdPJhlfGHvKM-AgXCsB1_ZhUISGp6UJY77QIT60YBNsSclCEX2X4mmucBx_JHf_APouU3GvCmwrV8y5gsb7EzxwBnIee7rjW_a8miqFD7ZboqAdAq7-R8J-f_Sc6cesCLY869Ar-4WyO366F1JveiRmHn2jWfySE83N6rFK4Q1HB2Kjh4qIjpnCqVG1w2ezADqWdYjc6FbzUY-sX00dQhSYs6TlqTZT7CJvpyUkCV6aFeCNinkjJorbwkReHOA0pTcQVT0Vhxd1Vt0MK5etx2E3Cp4Iq2BG4vxDnj0LGygMGlg7Ji2MsraX64gyPJQuxrmdEOTqhMiHnFjwa2xXB3JXSLfViG9EDAry3e8icQg5HpuxBJKSlFFh2xfyCxgrcsDaF7irn0DAGrVE5FxjX2R6NcBfYf8W_sX426o6HrjLzr_nDoek5v3MEn2HaHTtfz-uOB43g9Z-iNnzr4T27V6Q561-7A7Q9H7ngwuO71O5iGDEbFT3Y4zmfkp7_uT8Ws?type=png)](https://mermaid.live/edit#pako:eNrtVl1v2jAU_SuWn1qJIkIIhUirRGEgHrZVDCZ1QkImMcFqYme2s41V_e-7jhOapKnW91VIwbHP_b7HuY84ECHFPlb0R0Z5QGeMRJIkW45QSqRmAUsJ12i5mk3WiKh8sWyebhSV5tD8N88WkoRMbJbmvFw3MakUAVVql4aHnQhyVcvZXRM12e9Pp2-wIhE1kMnt7f09KjaaYEV1lu7gANYGmy_mJNBCnprY6VGKhMxuDbBcNzErqiWjP22cq8misrEWIn7ha83wlptzkmnBs2RvcyRpoJGM9hd9x-mg50ev27805wh9FpqimB40Eoc8tb7NPprH4peF2LJc3dw0U-ijTRoLEiISx-huNlcW30y0kaym1UeTTB_BYxYQME54iCKqkRYPlL9VQ4kIRZAloOriwGLaQeqBxfGOhUVwtWJetUagJeEKKsYEBzkrFguRog34B1FZPA2RgBqwhIpMW9BbvISwzh7uJFVZrNVF3eJlqe0NvipNdKbQh9IrxiPj1tlHq4rysC36NoVle6HCt9ezX-l0H301LyCTC5MYEdt-RrbKiFaTC6j0rEjJa_ZeGGwINI2UdPJhlfGHvKM-AgXCsB1_ZhUISGp6UJY77QIT60YBNsSclCEX2X4mmucBx_JHf_APouU3GvCmwrV8y5gsb7EzxwBnIee7rjW_a8miqFD7ZboqAdAq7-R8J-f_Sc6cesCLY869Ar-4WyO366F1JveiRmHn2jWfySE83N6rFK4Q1HB2Kjh4qIjpnCqVG1w2ezADqWdYjc6FbzUY-sX00dQhSYs6TlqTZT7CJvpyUkCV6aFeCNinkjJorbwkReHOA0pTcQVT0Vhxd1Vt0MK5etx2E3Cp4Iq2BG4vxDnj0LGygMGlg7Ji2MsraX64gyPJQuxrmdEOTqhMiHnFjwa2xXB3JXSLfViG9EDAry3e8icQg5HpuxBJKSlFFh2xfyCxgrcsDaF7irn0DAGrVE5FxjX2R6NcBfYf8W_sX426o6HrjLzr_nDoek5v3MEn2HaHTtfz-uOB43g9Z-iNnzr4T27V6Q561-7A7Q9H7ngwuO71O5iGDEbFT3Y4zmfkp7_uT8Ws)")

demo.launch(debug=True)