src/streamlit_app.py

from openai import OpenAI
import streamlit as st
import openai
import os 
import time
#from roles import *
import io
from pypdf import PdfReader  
#from langchain_community.document_loaders import PyPDFLoader
import tempfile
from RAG import load_graph,text_splitter
import torch 
from sentence_transformers import SentenceTransformer
import torch
import uuid 
import re
import requests
from cloudhands import CloudHandsPayment
from database_center import db_transaction
device='cuda' if torch.cuda.is_available() else 'cpu'
import os

# Explicitly override cache paths (matches Dockerfile ENV)
os.environ["HF_HOME"] = "/app/hf_cache"
os.environ["TRANSFORMERS_CACHE"] = "/app/hf_cache"

encoder = SentenceTransformer(
    "sentence-transformers/all-MiniLM-L6-v2",
    cache_folder="/app/hf_cache"
).to(device)


global chat_messages
chat_messages=[]
outputs=[]
# Set your OpenAI API key here or use environment variable
payment_key=os.getenv('Payment_Key')

def complete_payment():
    if st.session_state.token :
        chPay=st.session_state.chPay
        try:
            result = chPay.charge(
                charge=0.5,
                event_name="Sample cloudhands charge",
            )
            st.success(f"You payment is succeeded")
            st.session_state.transaction_id=result.transaction_id
            st.session_state.db_transaction.add({
            'id':str(uuid.uuid4()),
            'app':'app_title',
            'transaction-id':result.transaction_id,
            'price':0.5

            })
        except Exception as e:
            st.error(f"Charge failed: {e}")
    else:
        st.error('Please generate your Tokens.')




@st.dialog("Payment link")
def pay():
    chPay = st.session_state.chPay

    # Step 1: Show auth link only once
    auth_url = chPay.get_authorization_url()
    st.link_button("Authenticate", url=auth_url)

    # Step 2: User pastes the code
    code = st.text_input("Place your code")

    if st.button("Exchange Code"):
        try:
            token = chPay.exchange_code_for_token(code)
            st.session_state.token = token
            st.success("Code exchanged successfully! Token stored.")
        except Exception as e:
            st.error(f"Failed: {e}")

def embed_document(file_text):
    chunks=text_splitter.split_text(file_text)
    #embedded=[]
    embeddings=st.session_state.encoder.encode(chunks, convert_to_tensor=True, show_progress_bar=True)
    embeddings = embeddings.cpu().numpy()

    #embeddings=torch.concatenate(embedded).cpu().numpy()
    #embeddings=embeddings.cpu().numpy()
    #print(embedded)
    return embeddings,chunks


def embed_sentence(text):
    embeddings = st.session_state.encoder.encode([text], convert_to_tensor=True, show_progress_bar=True)  
    return embeddings.cpu().tolist() 



def stream_response():
    for char in extract_output(st.session_state.response).split(" "):
        yield char+" "
        time.sleep(0.1)  # Simulate a delay

def stream_thoughts():
    for char in extract_thinking(st.session_state.response).split(" "):
        yield char+" "
        time.sleep(0.1)  # Simulate a delay

def get_text(uploaded_file):
    # Save uploaded file to a temporary file
    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp_file:
        tmp_file.write(uploaded_file.read())
        tmp_path = tmp_file.name
    loader = PyPDFLoader(tmp_path)
    pages = loader.load()
    text = "\n".join([page.page_content for page in pages])
    return text


def respond_chat(text):

    url="https://8000-01k3gce7dwxsk16d7dd40n75xb.cloudspaces.litng.ai/predict"
    payload = { "user_prompt":text}
    headers = {"Content-Type": "application/json"}
    response = requests.post(url, data=payload)
        
    if response.status_code == 200:
        complete_payment()
        if st.session_state.transaction_id:
            return response.json()['output'][0]

def extract_thinking(text: str) -> str:
    """
    Extracts content inside <thinking>...</thinking> tags.
    Returns the first match or an empty string if not found.
    """
    match = re.search(r"<thinking>(.*?)</thinking>", text, re.DOTALL | re.IGNORECASE)
    return match.group(1).strip() if match else ""

def extract_output(text: str) -> str:
    """
    Extracts content inside <output>...</output> tags.
    Returns the first match or an empty string if not found.
    """
    match = re.search(r"<output>(.*?)</output>", text, re.DOTALL | re.IGNORECASE)
    return match.group(1).strip() if match else ""



# Dropdown for model selection
if 'doc_flag' not in st.session_state:
    st.session_state.doc_flag = False
if 'flag' not in st.session_state:
    st.session_state.flag = False
if 'encoder' not in st.session_state:
    st.session_state.encoder = encoder
if 'file_text' not in st.session_state:
    st.session_state.file_text = ""
if "chPay" not in st.session_state:
    st.session_state.chPay = CloudHandsPayment(
        author_key=payment_key
    )

if "token" not in st.session_state:
    st.session_state.token = None

if 'db_transaction' not in st.session_state:
    st.session_state.db_transaction = db_transaction
if 'embeddings' not in st.session_state:
    st.session_state.embeddings = None
if 'chunks' not in st.session_state:
    st.session_state.chunks = None
if 'response' not in st.session_state:
    st.session_state.response=''
# Sidebar document upload
st.sidebar.title("Uploading your document 📄")
uploaded_file = st.sidebar.file_uploader(
    "Upload your document 📄",
    type=["pdf"],
    label_visibility="collapsed"
)
upload_button=st.sidebar.button("Upload Document")
uploaded_file = st.sidebar.file_uploader(
    "Upload your PDF",
    type=["pdf"],
    key="pdf_uploader",
)

def extract_pdf_text_from_bytes(file_bytes: bytes) -> str:
    reader = PdfReader(io.BytesIO(file_bytes))
    pages_text = []
    for p in reader.pages:
        txt = p.extract_text() or ""
        pages_text.append(txt)
    return "\n".join(pages_text)

if uploaded_file is not None:
    with st.spinner("Reading & embedding your PDF..."):
        # Important: read bytes once on this rerun
        file_bytes = uploaded_file.read()
        # (Optional) if you ever re-use uploaded_file later, do: uploaded_file.seek(0)

        # Extract text purely in-memory (no /tmp files, no PyPDFLoader)
        file_text = extract_pdf_text_from_bytes(file_bytes)

        # Persist to session state
        st.session_state.file_text = file_text

        # Build embeddings (uses your existing text_splitter + encoder)
        chunks = text_splitter.split_text(file_text)
        embeddings = st.session_state.encoder.encode(
            chunks, convert_to_tensor=True, show_progress_bar=True
        ).cpu().numpy()

        st.session_state.embeddings = embeddings
        st.session_state.chunks = chunks
        st.session_state.doc_flag = True

    st.success(f"Loaded: {uploaded_file.name} — {len(st.session_state.chunks)} chunks")

st.sidebar.write("Before making the your faviorate charecter sound, authenicate your code")
Authenication=st.sidebar.button('Authenicate')
if Authenication:
    pay()

#subject=st.pills('Select your subject',list(roles.keys()),selection_mode='single')
st.title("Plaito")
st.write("Chat with our reasoning model and ask your questions. The model show you it's chain of thoughts and final answer.")
text=st.text_area("Ask your question:", height=100)
document_button=st.pills("Ask based on Documents", ['search'], selection_mode="single")
generate_button=st.button("Generate Response")
if generate_button:
    with st.spinner("Generating code..."):
        try:
            if document_button:
                graph=load_graph(st.session_state.embeddings,st.session_state.chunks)
                graph=graph.compile()
                initial_state = {
                    "embedded_query":embed_sentence(text),
                    "knowledge": [],
                    "summary": "",
                    "final_response": None,}
                final_state = graph.invoke(initial_state)
                updated_text = f"""
                Then respond to the client. Also follow the retrived information in the ##Summary section.
                ## Instructions:
                {text}
                ## Summary:
            {final_state['summary']}
                """
                if st.session_state.db_transaction:
                    response=respond_chat(updated_text)
                    st.session_state.response=response

            else: 
                if st.session_state.db_transaction:
                    response=respond_chat(text)
                    st.session_state.response=response
        except Exception as e:
                    st.error(f"Error during code generation: {e}")

    col1,col2=st.columns([2,1])
    with col2:
        st.write("### Thought Process")
        st.write_stream(stream_thoughts())
    with col1:
        st.write("### Response")
        st.write_stream(stream_response())