from sentence_transformers import SentenceTransformer import gradio as gr import os import json from bs4 import BeautifulSoup import requests from huggingface_hub import InferenceClient from langchain.vectorstores import Chroma # Required imports from sentence_transformers import SentenceTransformer from langchain.embeddings import HuggingFaceEmbeddings # Use Hugging Face wrapper for SentenceTransformers from langchain.document_loaders import DirectoryLoader, TextLoader from langchain.text_splitter import CharacterTextSplitter from langchain.schema import Document from langchain.vectorstores import Chroma import numpy as np from sklearn.manifold import TSNE import plotly.graph_objects as go from langchain.document_loaders import DirectoryLoader, TextLoader from langchain.text_splitter import CharacterTextSplitter from langchain.schema import Document import chromadb.utils.embedding_functions as embedding_functions from langchain.embeddings import HuggingFaceEmbeddings hf_token = os.getenv('HF_TOKEN') huggingface_ef = embedding_functions.HuggingFaceEmbeddingFunction( api_key=hf_token, model_name="sentence-transformers/all-MiniLM-L6-v2" ) embedding_model = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L6-v2') # Define global variables BOT_AVATAR = 'https://automatedstockmining.org/wp-content/uploads/2024/08/south-west-value-mining-logo.webp' #for the search vector database # Initialize Chroma vector store directory db_name2 = "search_checkvector_db" # Read in the text for processing health_check_text = '' with open('search_requirements.txt', 'r', encoding='utf-8') as search_text: search_requirements_text = search_text.read() # Split text into chunks search_splitter = CharacterTextSplitter(chunk_size=20, chunk_overlap=2) parts = search_splitter.split_text(search_requirements_text) search_documents = [Document(page_content=chunk) for chunk in chunks] # Initialize Chroma with documents and embeddings search_vectorstore = Chroma.from_documents( documents=search_documents, embedding=embedding_model, persist_directory=db_name ) # Initialize Chroma vector store directory db_name = "health_checkvector_db" # Read in the text for processing health_check_text = '' with open('healthcheck.txt', 'r', encoding='utf-8') as file: health_check_text = file.read() # Split text into chunks text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=200) chunks = text_splitter.split_text(health_check_text) # Convert chunks into Document objects documents = [Document(page_content=chunk) for chunk in chunks] # Initialize Chroma with documents and embeddings vectorstore = Chroma.from_documents( documents=documents, embedding=embedding_model, persist_directory=db_name ) client = InferenceClient(token=hf_token) custom_css = ''' .gradio-container { font-family: 'Roboto', sans-serif; } .main-header { text-align: center; color: #4a4a4a; margin-bottom: 2rem; } .tab-header { font-size: 1.2rem; font-weight: bold; margin-bottom: 1rem; } .custom-chatbot { border-radius: 10px; box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1); } .custom-button { background-color: #3498db; color: white; border: none; padding: 10px 20px; border-radius: 5px; cursor: pointer; transition: background-color 0.3s ease; } .custom-button:hover { background-color: #2980b9; } ''' def extract_text_from_webpage(html): soup = BeautifulSoup(html, "html.parser") for script in soup(["script", "style"]): script.decompose() visible_text = soup.get_text(separator=" ", strip=True) return visible_text def search(query): term = query max_chars_per_page = 8000 all_results = [] with requests.Session() as session: try: resp = session.get( url="https://www.google.com/search", headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0"}, params={"q": term, "num": 7}, timeout=5 ) resp.raise_for_status() soup = BeautifulSoup(resp.text, "html.parser") result_block = soup.find_all("div", attrs={"class": "g"}) for result in result_block: link = result.find("a", href=True) if link: link = link["href"] try: webpage = session.get(link, headers={"User-Agent": "Mozilla/5.0"}, timeout=5) webpage.raise_for_status() visible_text = extract_text_from_webpage(webpage.text) if len(visible_text) > max_chars_per_page: visible_text = visible_text[:max_chars_per_page] all_results.append({"link": link, "text": visible_text}) except requests.exceptions.RequestException as e: print(f"Failed to retrieve {link}: {e}") all_results.append({"link": link, "text": None}) except requests.exceptions.RequestException as e: print(f"Google search failed: {e}") return all_results def process_query(user_input, history): yield 'locating vectorstore 🛠️' docs = vectorstore.similarity_search(user_input, k=5) # Retrieve and concatenate results retrieved_texts = " ".join([doc.page_content for doc in docs]) #similarity search on searches searches = search_vectorstore.similarity_search(user_input, k=3) # Retrieve and concatenate results search_texts = " ".join([doc.page_content for doc in searches]) yield 'Preparing your request 🛠️' # Step 1: Generate a search term based on the user query stream_search = client.chat_completion( model="Qwen/Qwen2.5-72B-Instruct", messages=[{"role": "user", "content": f"Based on this chat history {history} the user's request '{user_input}', and this vector database {search_texts}, suggest a Google search term in a single line without specific dates; use 'this year', 'this month', etc. INCLUDE NOTHING IN YOUR RESPONSE EXCEPT THE RELEVANT SEARCH RESULT. EXAMPLE: USER: WHAT IS THE CURRENT PRICE OF COCA COLA STOCK. YOUR RESPONSE: WHAT IS THE CURRENT PRICE OF COCA COLA STOCK"}], max_tokens=400, stream=True ) # Collect the search term search_query = "" for chunk in stream_search: content = chunk.choices[0].delta.content or '' search_query += content # Step 2: Perform the web search with the generated term yield 'Searching the web for relevant information 🌐' search_results = search(search_query) # Format results as a JSON string for model input search_results_str = json.dumps(search_results) yield 'thinking...' # Step 3: Generate a response using the search results response = client.chat_completion( model="Qwen/Qwen2.5-72B-Instruct", messages=[{"role": "user", "content": f"Using the search results: {search_results_str} and chat history {history}, this vector database on health checks {retrieved_texts} answer the user's query '{user_input}' in a concise, precise way, using numerical data if available. ONLY GIVE ONE RESPONSE BACK, CONCISE OR DETAILED BASED ON THE USERS INPUT"}], max_tokens=3000, stream=True ) yield "Analyzing the data and getting ready to respond 📊" # Stream final response final_response = "" for chunk in response: content = chunk.choices[0].delta.content or '' final_response += content yield final_response theme = gr.themes.Citrus( primary_hue="blue", neutral_hue="slate", ) examples = [ ["whats the trending social sentiment like for Nvidia"], ["What's the latest news on Cisco Systems stock"], ["Analyze technical indicators for Adobe, are they presenting buy or sell signals"], ["Write me a smart sheet on the trending social sentiment and technical indicators for Nvidia"], ["What are the best stocks to buy this month"], ["What companies report earnings this week"], ["write me a health check on adobe"], ["Analyze the technical indicators for Apple"], ["Build an intrinsic value model for Apple"], ["Make a table of Apple's stock price for the last 3 days"], ["What is Apple's PE ratio and how does it compare to other companies in consumer electronics"], ["How did Salesforce perform in its last earnings?"], ["What is the average analyst price target for Nvidia"], ["What is the outlook for the stock market in 2025"], ["When does Nvidia next report earnings"], ["What are the latest products from Apple"], ["What is Tesla's current price-to-earnings ratio and how does it compare to other car manufacturers?"], ["List the top 5 performing stocks in the S&P 500 this month"], ["What is the dividend yield for Coca-Cola?"], ["Which companies in the tech sector are announcing dividends this month?"], ["Analyze the latest moving averages for Microsoft; are they indicating a trend reversal?"], ["What is the latest guidance on revenue for Meta?"], ["What is the current beta of Amazon stock and how does it compare to the industry average?"], ["What are the top-rated ETFs for technology exposure this quarter?"] ] chatbot = gr.Chatbot( label="IM.S", avatar_images=[None, BOT_AVATAR], show_copy_button=True, layout="panel", height=700 ) theme = gr.themes.Ocean() with gr.Blocks(theme=theme) as demo: with gr.Column(): gr.Markdown("## quantineuron.com: IM.analyst - Building the Future of Investing") with gr.Column(scale=3, min_width=600): chat_interface = gr.ChatInterface( fn=process_query, chatbot=chatbot, examples=examples ) with gr.Column(): gr.Markdown(''' **Disclaimer**: The information provided by IM.analyst is for educational and informational purposes only and does not constitute financial, investment, or professional advice. By using this service, you acknowledge and agree that all decisions you make based on the information provided are made at your own risk. Neither IM.analyst nor quantineuron.com is liable for any financial losses or damages resulting from reliance on information provided by this chatbot. By using IM.analyst, you agree to be bound by quantineuron.com’s [Terms of Service](https://quantineuron.com/disclaimer-statement/), [Terms and Conditions](https://quantineuron.com/terms-and-conditions/), [Data Protection and Privacy Policy](https://quantineuron.com/data-protection-and-privacy-policy/), [our discalimer statement](https://quantineuron.com/disclaimer-statement/) and this Disclaimer Statement. We recommend reviewing these documents carefully. Your continued use of this service confirms your acceptance of these terms and conditions, and it is your responsibility to stay informed of any updates or changes. **Important Note**: Investing in financial markets carries risk, and it is possible to lose some or all of the invested capital. Always consider seeking advice from a qualified financial advisor. ''') demo.launch()