Jashan1's picture
Update app.py (#2)
2ea75a2 verified
import os
import io
import requests
import streamlit as st
from openai import OpenAI
from PyPDF2 import PdfReader
import urllib.parse
from dotenv import load_dotenv
from openai import OpenAI
from io import BytesIO
from streamlit_extras.colored_header import colored_header
from streamlit_extras.add_vertical_space import add_vertical_space
from streamlit_extras.switch_page_button import switch_page
import json
import pandas as pd
from st_aggrid import AgGrid, GridOptionsBuilder, GridUpdateMode, DataReturnMode
import time
import random
import aiohttp
import asyncio
from PyPDF2 import PdfWriter
load_dotenv()
# ---------------------- Configuration ----------------------
st.set_page_config(page_title="Building Regulations Chatbot", layout="wide", initial_sidebar_state="expanded")
# Load environment variables from .env file
load_dotenv()
# Set OpenAI API key
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
# ---------------------- Session State Initialization ----------------------
if 'pdf_contents' not in st.session_state:
st.session_state.pdf_contents = []
if 'chat_history' not in st.session_state:
st.session_state.chat_history = []
if 'processed_pdfs' not in st.session_state:
st.session_state.processed_pdfs = False
if 'id_counter' not in st.session_state:
st.session_state.id_counter = 0
if 'assistant_id' not in st.session_state:
st.session_state.assistant_id = None
if 'thread_id' not in st.session_state:
st.session_state.thread_id = None
if 'file_ids' not in st.session_state:
st.session_state.file_ids = []
# ---------------------- Helper Functions ----------------------
def get_vector_stores():
try:
vector_stores = client.beta.vector_stores.list()
return vector_stores
except Exception as e:
return f"Error retrieving vector stores: {str(e)}"
def fetch_pdfs(city_code):
url = f"http://91.203.213.50:5000/oereblex/{city_code}"
response = requests.get(url)
if response.status_code == 200:
data = response.json()
print("First data:", data.get('data', [])[0] if data.get('data') else None)
return data.get('data', [])
else:
st.error(f"Failed to fetch PDFs for city code {city_code}")
return None
def download_pdf(url, doc_title):
# Add 'https://' scheme if it's missing
if not url.startswith(('http://', 'https://')):
url = 'https://' + url
try:
response = requests.get(url)
response.raise_for_status() # Raise an exception for bad status codes
# Sanitize doc_title to create a valid filename
sanitized_title = ''.join(c for c in doc_title if c.isalnum() or c in (' ', '_', '-')).rstrip()
sanitized_title = sanitized_title.replace(' ', '_')
filename = f"{sanitized_title}.pdf"
# Ensure filename is unique by appending the id_counter if necessary
if os.path.exists(filename):
filename = f"{sanitized_title}_{st.session_state.id_counter}.pdf"
st.session_state.id_counter += 1
# Save the PDF content to a file
with open(filename, 'wb') as f:
f.write(response.content)
return filename
except requests.RequestException as e:
st.error(f"Failed to download PDF from {url}. Error: {str(e)}")
return None
# Helper function to upload file to OpenAI
def upload_file_to_openai(file_path):
try:
file = client.files.create(
file=open(file_path, 'rb'),
purpose='assistants'
)
return file.id
except Exception as e:
st.error(f"Failed to upload file {file_path}. Error: {str(e)}")
return None
def create_assistant():
assistant = client.beta.assistants.create(
name="Building Regulations Assistant",
instructions="You are an expert on building regulations. Use the provided documents to answer questions accurately.",
model="gpt-4o-mini",
tools=[{"type": "file_search"}]
)
st.session_state.assistant_id = assistant.id
return assistant.id
def format_response(response, citations):
"""Format the response with proper markdown structure."""
parts = ["", response]
if citations:
parts.extend(["### Citations"])
parts.extend(f"- {citation}" for citation in citations)
return "\n\n".join(parts)
def response_generator(response, citations):
"""Generator for streaming response with structured output."""
# First yield the response heade
time.sleep(0.1)
# Yield the main response word by word
words = response.split()
for i, word in enumerate(words):
yield word + " "
# Add natural pauses at punctuation
if word.endswith(('.', '!', '?', ':')):
time.sleep(0.1)
else:
time.sleep(0.05)
# If there are citations, yield them with proper formatting
if citations:
# Add some spacing before citations
yield "\n\n### Citations\n\n"
time.sleep(0.1)
for citation in citations:
yield f"- {citation}\n"
time.sleep(0.05)
def chat_with_assistant(file_ids, user_message):
print("----- Starting chat_with_assistant -----")
print("Received file_ids:", file_ids)
print("Received user_message:", user_message)
# Create attachments for each file_id
attachments = [{"file_id": file_id, "tools": [{"type": "file_search"}]} for file_id in file_ids]
print("Attachments created:", attachments)
if st.session_state.thread_id is None:
print("No existing thread_id found. Creating a new thread.")
thread = client.beta.threads.create(
messages=[
{
"role": "user",
"content": user_message,
"attachments": attachments,
}
]
)
st.session_state.thread_id = thread.id
print("New thread created with id:", st.session_state.thread_id)
else:
print(f"Existing thread_id found: {st.session_state.thread_id}. Adding message to the thread.")
message = client.beta.threads.messages.create(
thread_id=st.session_state.thread_id,
role="user",
content=user_message,
attachments=attachments
)
print("Message added to thread with id:", message.id)
try:
thread = client.beta.threads.retrieve(thread_id=st.session_state.thread_id)
print("Retrieved thread:", thread)
except Exception as e:
print(f"Error retrieving thread with id {st.session_state.thread_id}: {e}")
return "An error occurred while processing your request.", []
try:
run = client.beta.threads.runs.create_and_poll(
thread_id=thread.id, assistant_id=st.session_state.assistant_id
)
print("Run created and polled:", run)
except Exception as e:
print("Error during run creation and polling:", e)
return "An error occurred while processing your request.", []
try:
messages = list(client.beta.threads.messages.list(thread_id=thread.id, run_id=run.id))
print("Retrieved messages:", messages)
except Exception as e:
print("Error retrieving messages:", e)
return "An error occurred while retrieving messages.", []
# Process the first message content
if messages and messages[0].content:
message_content = messages[0].content[0].text
print("Raw message content:", message_content)
annotations = message_content.annotations
citations = []
seen_citations = set()
# Process annotations and citations
for index, annotation in enumerate(annotations):
message_content.value = message_content.value.replace(annotation.text, f"[{index}]")
if file_citation := getattr(annotation, "file_citation", None):
try:
cited_file = client.files.retrieve(file_citation.file_id)
citation_entry = f"[{index}] {cited_file.filename}"
if citation_entry not in seen_citations:
citations.append(citation_entry)
seen_citations.add(citation_entry)
except Exception as e:
print(f"Error retrieving cited file for annotation {index}: {e}")
# Create a container for the response with proper styling
response_container = st.container()
with response_container:
message_placeholder = st.empty()
streaming_content = ""
# Stream the response with structure
for chunk in response_generator(message_content.value, citations):
streaming_content += chunk
# Use markdown for proper formatting during streaming
message_placeholder.markdown(streaming_content + "▌")
# Final formatted response
final_formatted_response = format_response(message_content.value, citations)
message_placeholder.markdown(final_formatted_response)
return final_formatted_response, citations
else:
return "No response received from the assistant.", []
# ---------------------- Streamlit App ----------------------
# ---------------------- Custom CSS Injection ----------------------
# Inject custom CSS to style chat messages
st.markdown("""
<style>
/* Style for the chat container */
.chat-container {
display: flex;
flex-direction: column;
gap: 1.5rem;
}
/* Style for individual chat messages */
.chat-message {
margin-bottom: 1.5rem;
}
/* Style for user messages */
.chat-message.user > div:first-child {
color: #1E90FF; /* Dodger Blue for "You" */
font-weight: bold;
margin-bottom: 0.5rem;
}
/* Style for assistant messages */
.chat-message.assistant > div:first-child {
color: #32CD32; /* Lime Green for "Assistant" */
font-weight: bold;
margin-bottom: 0.5rem;
}
/* Style for the message content */
.message-content {
padding: 1rem;
border-radius: 0.5rem;
line-height: 1.5;
}
.message-content h3 {
color: #444;
margin-top: 1rem;
margin-bottom: 0.5rem;
font-size: 1.1rem;
}
.message-content ul {
margin-top: 0.5rem;
margin-bottom: 0.5rem;
padding-left: 1.5rem;
}
.message-content li {
margin-bottom: 0.25rem;
}
</style>
""", unsafe_allow_html=True)
page = st.sidebar.selectbox("Choose a page", ["Documents", "Home", "Admin"])
if page == "Home":
st.title("Building Regulations Chatbot", anchor=False)
# Sidebar improvements
with st.sidebar:
colored_header("Selected Documents", description="Documents for chat")
if 'selected_pdfs' in st.session_state and not st.session_state.selected_pdfs.empty:
for _, pdf in st.session_state.selected_pdfs.iterrows():
st.write(f"- {pdf['Doc Title']}")
else:
st.write("No documents selected. Please go to the Documents page.")
# Main chat area improvements
colored_header("Chat", description="Ask questions about building regulations")
# Chat container with custom CSS class
st.markdown('<div class="chat-container" id="chat-container">', unsafe_allow_html=True)
for chat in st.session_state.chat_history:
with st.container():
if chat['role'] == 'user':
st.markdown(f"""
<div class="chat-message user">
<div><strong>You</strong></div>
<div class="message-content">{chat['content']}</div>
</div>
""", unsafe_allow_html=True)
else:
st.markdown(f"""
<div class="chat-message assistant">
<div><strong>Assistant</strong></div>
<div class="message-content">{chat['content']}</div>
</div>
""", unsafe_allow_html=True)
st.markdown('</div>', unsafe_allow_html=True)
# Inject JavaScript to auto-scroll the chat container
st.markdown("""
<script>
const chatContainer = document.getElementById('chat-container');
if (chatContainer) {
chatContainer.scrollTop = chatContainer.scrollHeight;
}
</script>
""", unsafe_allow_html=True)
# Chat input improvements
with st.form("chat_form", clear_on_submit=True):
user_input = st.text_area("Ask a question about building regulations...", height=100)
col1, col2 = st.columns([3, 1])
with col2:
submit = st.form_submit_button("Send", use_container_width=True)
if submit and user_input.strip() != "":
# Add user message to chat history
st.session_state.chat_history.append({"role": "user", "content": user_input})
if not st.session_state.file_ids:
st.error("Please process PDFs first.")
else:
with st.spinner("Generating response..."):
try:
response, citations = chat_with_assistant(st.session_state.file_ids, user_input)
# The response is already formatted, so we can add it directly to chat history
st.session_state.chat_history.append({
"role": "assistant",
"content": response
})
except Exception as e:
st.error(f"Error generating response: {str(e)}")
# Rerun the app to update the chat display
st.rerun()
# Footer improvements
add_vertical_space(2)
st.markdown("---")
col1, col2 = st.columns(2)
with col1:
st.caption("Powered by OpenAI GPT-4 and Pinecone")
with col2:
st.caption("© 2023 Your Company Name")
elif page == "Documents":
st.title("Document Selection")
city_code_input = st.text_input("Enter city code:", key="city_code_input")
load_documents_button = st.button("Load Documents", key="load_documents_button")
if load_documents_button and city_code_input:
with st.spinner("Fetching PDFs..."):
pdfs = fetch_pdfs(city_code_input)
if pdfs:
st.session_state.available_pdfs = pdfs
st.success(f"Found {len(pdfs)} PDFs")
else:
st.error("No PDFs found")
if 'available_pdfs' in st.session_state:
st.write(f"Total PDFs: {len(st.session_state.available_pdfs)}")
# Create a DataFrame from the available PDFs
df = pd.DataFrame(st.session_state.available_pdfs)
# Select and rename only the specified columns
df = df[['municipality', 'abbreviation', 'doc_title', 'file_title', 'file_href', 'enactment_date', 'prio']]
df = df.rename(columns={
"municipality": "Municipality",
"abbreviation": "Abbreviation",
"doc_title": "Doc Title",
"file_title": "File Title",
"file_href": "File Href",
"enactment_date": "Enactment Date",
"prio": "Prio"
})
# Add a checkbox column to the DataFrame at the beginning
df.insert(0, "Select", False)
# Configure grid options
gb = GridOptionsBuilder.from_dataframe(df)
gb.configure_default_column(enablePivot=True, enableValue=True, enableRowGroup=True)
gb.configure_column("Select", header_name="Select", cellRenderer='checkboxRenderer')
gb.configure_column("File Href", cellRenderer='linkRenderer')
gb.configure_selection(selection_mode="multiple", use_checkbox=True)
gb.configure_side_bar()
gridOptions = gb.build()
# Display the AgGrid
grid_response = AgGrid(
df,
gridOptions=gridOptions,
enable_enterprise_modules=True,
update_mode=GridUpdateMode.MODEL_CHANGED,
data_return_mode=DataReturnMode.FILTERED_AND_SORTED,
fit_columns_on_grid_load=False,
)
# Get the selected rows
selected_rows = grid_response['selected_rows']
# Debug: Print the structure of selected_rows
st.write("Debug - Selected Rows Structure:", selected_rows)
if st.button("Process Selected PDFs"):
if len(selected_rows) > 0: # Check if there are any selected rows
# Convert selected_rows to a DataFrame
st.session_state.selected_pdfs = pd.DataFrame(selected_rows)
st.session_state.assistant_id = create_assistant()
with st.spinner("Processing PDFs and creating/updating assistant..."):
file_ids = []
for _, pdf in st.session_state.selected_pdfs.iterrows():
# Debug: Print each pdf item
st.write("Debug - PDF item:", pdf)
file_href = pdf['File Href']
doc_title = pdf['Doc Title']
# Pass doc_title to download_pdf
file_name = download_pdf(file_href, doc_title)
if file_name:
file_path = f"./{file_name}"
file_id = upload_file_to_openai(file_path)
if file_id:
file_ids.append(file_id)
else:
st.warning(f"Failed to upload {doc_title}. Skipping this file.")
else:
st.warning(f"Failed to download {doc_title}. Skipping this file.")
st.session_state.file_ids = file_ids
st.success("PDFs processed successfully. You can now chat on the Home page.")
else:
st.warning("Select at least one PDF.")
elif page == "Admin":
st.title("Admin Panel")
st.header("Vector Stores Information")
vector_stores = get_vector_stores()
json_vector_stores = json.dumps([vs.model_dump() for vs in vector_stores])
st.write(json_vector_stores)
# Add a button to go back to the main page