from flask import Flask, request, jsonify import requests from bs4 import BeautifulSoup import yfinance as yf import pandas as pd from datetime import datetime, timedelta import logging from concurrent.futures import ThreadPoolExecutor, as_completed from langchain_google_genai import ChatGoogleGenerativeAI from config import Config import numpy as np from typing import Optional, Tuple, List, Dict from rag import get_answer import time from tenacity import retry, stop_after_attempt, wait_exponential # Initialize Flask app app = Flask(__name__) # Set up logging logging.basicConfig(level=logging.DEBUG, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s', handlers=[logging.FileHandler("app.log"), logging.StreamHandler()]) logger = logging.getLogger(__name__) # Initialize the Gemini model llm = ChatGoogleGenerativeAI(api_key=Config.GEMINI_API_KEY, model="gemini-1.5-flash-latest", temperature=0.5) # Configuration for Google Custom Search API GOOGLE_API_KEY = Config.GOOGLE_API_KEY SEARCH_ENGINE_ID = Config.SEARCH_ENGINE_ID @retry(stop=stop_after_attempt(5), wait=wait_exponential(multiplier=1, min=2, max=8), reraise=True) def invoke_llm(prompt): return llm.invoke(prompt) class DataSummarizer: def __init__(self): pass def google_search(self, query: str) -> Optional[str]: start_time = time.time() try: url = "https://www.googleapis.com/customsearch/v1" params = { 'key': GOOGLE_API_KEY, 'cx': SEARCH_ENGINE_ID, 'q': query } response = requests.get(url, params=params) response.raise_for_status() search_results = response.json() logger.info("google_search took %.2f seconds", time.time() - start_time) # Summarize the search results using Gemini items = search_results.get('items', []) content = "\n\n".join([f"{item.get('title', '')}\n{item.get('snippet', '')}" for item in items]) prompt = f"Summarize the following search results:\n\n{content}" summary_response = invoke_llm(prompt) return summary_response.content.strip() except Exception as e: logger.error(f"Error during Google Search API request: {e}") return None def extract_content_from_item(self, item: Dict) -> Optional[str]: try: snippet = item.get('snippet', '') title = item.get('title', '') return f"{title}\n{snippet}" except Exception as e: logger.error(f"Error extracting content from item: {e}") return None def calculate_moving_average(self, df: pd.DataFrame, window: int = 20) -> Optional[pd.Series]: start_time = time.time() try: result = df['close'].rolling(window=window).mean() logger.info("calculate_moving_average took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating moving average: {e}") return None def calculate_rsi(self, df: pd.DataFrame, window: int = 14) -> Optional[pd.Series]: start_time = time.time() try: delta = df['close'].diff() gain = delta.where(delta > 0, 0).rolling(window=window).mean() loss = -delta.where(delta < 0, 0).rolling(window=window).mean() rs = gain / loss result = 100 - (100 / (1 + rs)) logger.info("calculate_rsi took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating RSI: {e}") return None def calculate_ema(self, df: pd.DataFrame, window: int = 20) -> Optional[pd.Series]: start_time = time.time() try: result = df['close'].ewm(span=window, adjust=False).mean() logger.info("calculate_ema took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating EMA: {e}") return None def calculate_bollinger_bands(self, df: pd.DataFrame, window: int = 20) -> Optional[pd.DataFrame]: start_time = time.time() try: ma = df['close'].rolling(window=window).mean() std = df['close'].rolling(window=window).std() upper_band = ma + (std * 2) lower_band = ma - (std * 2) result = pd.DataFrame({'MA': ma, 'Upper Band': upper_band, 'Lower Band': lower_band}) logger.info("calculate_bollinger_bands took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating Bollinger Bands: {e}") return None def calculate_macd(self, df: pd.DataFrame, short_window: int = 12, long_window: int = 26, signal_window: int = 9) -> \ Optional[pd.DataFrame]: start_time = time.time() try: short_ema = df['close'].ewm(span=short_window, adjust=False).mean() long_ema = df['close'].ewm(span=long_window, adjust=False).mean() macd = short_ema - long_ema signal = macd.ewm(span=signal_window, adjust=False).mean() result = pd.DataFrame({'MACD': macd, 'Signal Line': signal}) logger.info("calculate_macd took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating MACD: {e}") return None def calculate_volatility(self, df: pd.DataFrame, window: int = 20) -> Optional[pd.Series]: start_time = time.time() try: log_returns = np.log(df['close'] / df['close'].shift(1)) result = log_returns.rolling(window=window).std() * np.sqrt(window) logger.info("calculate_volatility took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating volatility: {e}") return None def calculate_atr(self, df: pd.DataFrame, window: int = 14) -> Optional[pd.Series]: start_time = time.time() try: high_low = df['high'] - df['low'] high_close = np.abs(df['high'] - df['close'].shift()) low_close = np.abs(df['low'] - df['close'].shift()) true_range = pd.concat([high_low, high_close, low_close], axis=1).max(axis=1) result = true_range.rolling(window=window).mean() logger.info("calculate_atr took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating ATR: {e}") return None def calculate_obv(self, df: pd.DataFrame) -> Optional[pd.Series]: start_time = time.time() try: result = (np.sign(df['close'].diff()) * df['volume']).fillna(0).cumsum() logger.info("calculate_obv took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating OBV: {e}") return None def calculate_yearly_summary(self, df: pd.DataFrame) -> Optional[pd.DataFrame]: start_time = time.time() try: df['year'] = pd.to_datetime(df['date']).dt.year yearly_summary = df.groupby('year').agg({ 'close': ['mean', 'max', 'min'], 'volume': 'sum' }) yearly_summary.columns = ['_'.join(col) for col in yearly_summary.columns] logger.info("calculate_yearly_summary took %.2f seconds", time.time() - start_time) return yearly_summary except Exception as e: logger.error(f"Error calculating yearly summary: {e}") return None def get_full_last_year(self, df: pd.DataFrame) -> Optional[pd.DataFrame]: start_time = time.time() try: today = datetime.today().date() last_year_start = datetime(today.year - 1, 1, 1).date() last_year_end = datetime(today.year - 1, 12, 31).date() mask = (df['date'] >= last_year_start) & (df['date'] <= last_year_end) result = df.loc[mask] logger.info("get_full_last_year took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error filtering data for the last year: {e}") return None def calculate_ytd_performance(self, df: pd.DataFrame) -> Optional[float]: start_time = time.time() try: today = datetime.today().date() year_start = datetime(today.year, 1, 1).date() mask = (df['date'] >= year_start) & (df['date'] <= today) ytd_data = df.loc[mask] opening_price = ytd_data.iloc[0]['open'] closing_price = ytd_data.iloc[-1]['close'] result = ((closing_price - opening_price) / opening_price) * 100 logger.info("calculate_ytd_performance took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating YTD performance: {e}") return None def calculate_pe_ratio(self, current_price: float, eps: float) -> Optional[float]: start_time = time.time() try: if eps == 0: raise ValueError("EPS cannot be zero for P/E ratio calculation.") result = current_price / eps logger.info("calculate_pe_ratio took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error calculating P/E ratio: {e}") return None def fetch_google_snippet(self, query: str) -> Optional[str]: try: search_url = f"https://www.google.com/search?q={query}" headers = { "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.36" } response = requests.get(search_url, headers=headers) soup = BeautifulSoup(response.text, 'html.parser') snippet_classes = [ 'BNeawe iBp4i AP7Wnd', 'BNeawe s3v9rd AP7Wnd', 'BVG0Nb', 'kno-rdesc' ] snippet = None for cls in snippet_classes: snippet = soup.find('div', class_=cls) if snippet: break return snippet.get_text() if snippet else "Snippet not found." except Exception as e: logger.error(f"Error fetching Google snippet: {e}") return None def extract_ticker_from_response(response: str) -> Optional[str]: start_time = time.time() try: if "is **" in response and "**." in response: result = response.split("is **")[1].split("**.")[0].strip() logger.info("extract_ticker_from_response took %.2f seconds", time.time() - start_time) return result result = response.strip() logger.info("extract_ticker_from_response took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Error extracting ticker from response: {e}") return None def detect_translate_entity_and_ticker(query: str) -> Tuple[Optional[str], Optional[str], Optional[str], Optional[str]]: try: start_time = time.time() # Step 1: Detect Language prompt = f"Detect the language for the following text: {query}" response = invoke_llm(prompt) detected_language = response.content.strip() logger.info(f"Language detected: {detected_language}") # Step 2: Translate to English (if necessary) translated_query = query if detected_language != "English": prompt = f"Translate the following text to English: {query}" response = invoke_llm(prompt) translated_query = response.content.strip() logger.info(f"Translation completed: {translated_query}") print(f"Translation: {translated_query}") # Step 3: Detect Entity prompt = f"Detect the entity in the following text that is a company name: {translated_query}" response = invoke_llm(prompt) detected_entity = response.content.strip() logger.info(f"Entity detected: {detected_entity}") print(f"Entity: {detected_entity}") if not detected_entity: logger.error("No entity detected") return detected_language, None, translated_query, None # Step 4: Get Stock Ticker prompt = f"What is the stock ticker symbol for the company {detected_entity}?" response = invoke_llm(prompt) stock_ticker = extract_ticker_from_response(response.content.strip()) if not stock_ticker: logger.error("No stock ticker detected") return detected_language, detected_entity, translated_query, None logger.info("detect_translate_entity_and_ticker took %.2f seconds", time.time() - start_time) return detected_language, detected_entity, translated_query, stock_ticker except Exception as e: logger.error(f"Error in detecting, translating, or extracting entity and ticker: {e}") return None, None, None, None def fetch_stock_data_yahoo(symbol: str) -> pd.DataFrame: start_time = time.time() try: stock = yf.Ticker(symbol) logger.info(f"Fetching data for symbol: {symbol}") end_date = datetime.now() start_date = end_date - timedelta(days=3 * 365) historical_data = stock.history(start=start_date, end=end_date) if historical_data.empty: raise ValueError(f"No historical data found for symbol: {symbol}") historical_data = historical_data.rename( columns={"Open": "open", "High": "high", "Low": "low", "Close": "close", "Volume": "volume"} ) historical_data.reset_index(inplace=True) historical_data['date'] = historical_data['Date'].dt.date historical_data = historical_data.drop(columns=['Date']) historical_data = historical_data[['date', 'open', 'high', 'low', 'close', 'volume']] if 'close' not in historical_data.columns: raise KeyError("The historical data must contain a 'close' column.") logger.info("fetch_stock_data_yahoo took %.2f seconds", time.time() - start_time) return historical_data except Exception as e: logger.error(f"Failed to fetch stock data for {symbol} from Yahoo Finance: {e}") return pd.DataFrame() def fetch_current_stock_price(symbol: str) -> Optional[float]: start_time = time.time() try: stock = yf.Ticker(symbol) result = stock.info['currentPrice'] logger.info("fetch_current_stock_price took %.2f seconds", time.time() - start_time) return result except Exception as e: logger.error(f"Failed to fetch current stock price for {symbol}: {e}") return None def format_stock_data_for_gemini(stock_data: pd.DataFrame) -> str: start_time = time.time() try: if stock_data.empty: return "No historical data available." formatted_data = "Historical stock data for the last three years:\n\n" formatted_data += "Date | Open | High | Low | Close | Volume\n" formatted_data += "------------------------------------------------------\n" for index, row in stock_data.iterrows(): formatted_data += f"{row['date']} | {row['open']:.2f} | {row['high']:.2f} | {row['low']:.2f} | {row['close']:.2f} | {int(row['volume'])}\n" logger.info("format_stock_data_for_gemini took %.2f seconds", time.time() - start_time) return formatted_data except Exception as e: logger.error(f"Error formatting stock data for Gemini: {e}") return "Error formatting stock data." def fetch_company_info_yahoo(symbol: str) -> Dict: start_time = time.time() try: if not symbol: return {"error": "Invalid symbol"} stock = yf.Ticker(symbol) company_info = stock.info logger.info("fetch_company_info_yahoo took %.2f seconds", time.time() - start_time) return { "name": company_info.get("longName", "N/A"), "sector": company_info.get("sector", "N/A"), "industry": company_info.get("industry", "N/A"), "marketCap": company_info.get("marketCap", "N/A"), "summary": company_info.get("longBusinessSummary", "N/A"), "website": company_info.get("website", "N/A"), "address": company_info.get("address1", "N/A"), "city": company_info.get("city", "N/A"), "state": company_info.get("state", "N/A"), "country": company_info.get("country", "N/A"), "phone": company_info.get("phone", "N/A") } except Exception as e: logger.error(f"Error fetching company info for {symbol}: {e}") return {"error": str(e)} def format_company_info_for_gemini(company_info: Dict) -> str: start_time = time.time() try: if "error" in company_info: return f"Error fetching company info: {company_info['error']}" formatted_info = (f"\nCompany Information:\n" f"Name: {company_info['name']}\n" f"Sector: {company_info['sector']}\n" f"Industry: {company_info['industry']}\n" f"Market Cap: {company_info['marketCap']}\n" f"Summary: {company_info['summary']}\n" f"Website: {company_info['website']}\n" f"Address: {company_info['address']}, {company_info['city']}, {company_info['state']}, {company_info['country']}\n" f"Phone: {company_info['phone']}\n") logger.info("format_company_info_for_gemini took %.2f seconds", time.time() - start_time) return formatted_info except Exception as e: logger.error(f"Error formatting company info for Gemini: {e}") return "Error formatting company info." def fetch_company_news_yahoo(symbol: str) -> List[Dict]: start_time = time.time() try: stock = yf.Ticker(symbol) news = stock.news if not news: raise ValueError(f"No news found for symbol: {symbol}") logger.info("fetch_company_news_yahoo took %.2f seconds", time.time() - start_time) return news except Exception as e: logger.error(f"Failed to fetch news for {symbol} from Yahoo Finance: {e}") return [] def format_company_news_for_gemini(news: List[Dict]) -> str: start_time = time.time() try: if not news: return "No news available." formatted_news = "Latest company news:\n\n" for article in news: formatted_news += (f"Title: {article['title']}\n" f"Publisher: {article['publisher']}\n" f"Link: {article['link']}\n" f"Published: {article['providerPublishTime']}\n\n") logger.info("format_company_news_for_gemini took %.2f seconds", time.time() - start_time) return formatted_news except Exception as e: logger.error(f"Error formatting company news for Gemini: {e}") return "Error formatting company news." def send_to_gemini_for_summarization(content: str) -> str: start_time = time.time() try: unified_content = " ".join(content) prompt = f"Summarize the main points of this article.\n\n{unified_content}" response = invoke_llm(prompt) logger.info("send_to_gemini_for_summarization took %.2f seconds", time.time() - start_time) return response.content.strip() except Exception as e: logger.error(f"Error sending content to Gemini for summarization: {e}") return "Error summarizing content." def answer_question_with_data(question: str, data: Dict) -> str: start_time = time.time() try: data_str = "" for key, value in data.items(): data_str += f"{key}:\n{value}\n\n" prompt = (f"You are a financial advisor. Begin your answer and only give the answer after.\n" f"Using the following data, answer this question: {question}\n\nData:\n{data_str}\n" f"Make your answer in the best form and professional.\n" f"Don't say anything about the source of the data.\n" f"If you don't have the data to answer, say this data is not available yet. If the data is not available in the stock history data, say this was a weekend and there is no data for it.") response = invoke_llm(prompt) logger.info("answer_question_with_data took %.2f seconds", time.time() - start_time) return response.content.strip() except Exception as e: logger.error(f"Error answering question with data: {e}") return "Error answering question." def calculate_metrics(stock_data: pd.DataFrame, summarizer: DataSummarizer, company_info: Dict) -> Dict[str, str]: start_time = time.time() try: moving_average = summarizer.calculate_moving_average(stock_data) rsi = summarizer.calculate_rsi(stock_data) ema = summarizer.calculate_ema(stock_data) bollinger_bands = summarizer.calculate_bollinger_bands(stock_data) macd = summarizer.calculate_macd(stock_data) volatility = summarizer.calculate_volatility(stock_data) atr = summarizer.calculate_atr(stock_data) obv = summarizer.calculate_obv(stock_data) yearly_summary = summarizer.calculate_yearly_summary(stock_data) ytd_performance = summarizer.calculate_ytd_performance(stock_data) eps = company_info.get('trailingEps', None) if eps: current_price = stock_data.iloc[-1]['close'] pe_ratio = summarizer.calculate_pe_ratio(current_price, eps) formatted_metrics = { "Moving Average": moving_average.to_string(), "RSI": rsi.to_string(), "EMA": ema.to_string(), "Bollinger Bands": bollinger_bands.to_string(), "MACD": macd.to_string(), "Volatility": volatility.to_string(), "ATR": atr.to_string(), "OBV": obv.to_string(), "Yearly Summary": yearly_summary.to_string(), "YTD Performance": f"{ytd_performance:.2f}%", "P/E Ratio": f"{pe_ratio:.2f}" } else: formatted_metrics = { "Moving Average": moving_average.to_string(), "RSI": rsi.to_string(), "EMA": ema.to_string(), "Bollinger Bands": bollinger_bands.to_string(), "MACD": macd.to_string(), "Volatility": volatility.to_string(), "ATR": atr.to_string(), "OBV": obv.to_string(), "Yearly Summary": yearly_summary.to_string(), "YTD Performance": f"{ytd_performance:.2f}%" } logger.info("calculate_metrics took %.2f seconds", time.time() - start_time) return formatted_metrics except Exception as e: logger.error(f"Error calculating metrics: {e}") return {"Error": "Error calculating metrics"} def prepare_data(formatted_stock_data: str, formatted_company_info: str, formatted_company_news: str, google_results: str, formatted_metrics: Dict[str, str], google_snippet: str, rag_response: str) -> \ Dict[str, str]: start_time = time.time() collected_data = { "Formatted Stock Data": formatted_stock_data, "Formatted Company Info": formatted_company_info, "Formatted Company News": formatted_company_news, "Google Search Results": google_results, "Google Snippet": google_snippet, "RAG Response": rag_response, "Calculations": formatted_metrics } collected_data.update(formatted_metrics) logger.info("prepare_data took %.2f seconds", time.time() - start_time) return collected_data @app.route('/ask', methods=['POST']) def ask(): try: user_input = request.json.get('question') logger.info(f"Received question: {user_input}") summarizer = DataSummarizer() # Detect language, entity, translation, and stock ticker language, entity, translation, stock_ticker = detect_translate_entity_and_ticker(user_input) logger.info(f"Detected Language: {language}, Entity: {entity}, Translation: {translation}, Stock Ticker: {stock_ticker}") if entity and stock_ticker: with ThreadPoolExecutor() as executor: futures = { executor.submit(fetch_stock_data_yahoo, stock_ticker): "stock_data", executor.submit(fetch_company_info_yahoo, stock_ticker): "company_info", executor.submit(fetch_company_news_yahoo, stock_ticker): "company_news", executor.submit(fetch_current_stock_price, stock_ticker): "current_stock_price", executor.submit(get_answer, user_input): "rag_response", executor.submit(summarizer.google_search, user_input): "google_results", executor.submit(summarizer.fetch_google_snippet, user_input): "google_snippet" } results = {futures[future]: future.result() for future in as_completed(futures)} stock_data = results.get("stock_data", pd.DataFrame()) formatted_stock_data = format_stock_data_for_gemini(stock_data) if not stock_data.empty else "No historical data available." company_info = results.get("company_info", {}) formatted_company_info = format_company_info_for_gemini(company_info) if company_info else "No company info available." company_news = results.get("company_news", []) formatted_company_news = format_company_news_for_gemini(company_news) if company_news else "No news available." current_stock_price = results.get("current_stock_price", None) formatted_metrics = calculate_metrics(stock_data, summarizer, company_info) if not stock_data.empty else {"Error": "No stock data for metrics"} google_results = results.get("google_results", "No additional news found through Google Search.") google_snippet = results.get("google_snippet", "Snippet not found.") rag_response = results.get("rag_response", "No response from RAG.") collected_data = prepare_data(formatted_stock_data, formatted_company_info, formatted_company_news, google_results, formatted_metrics, google_snippet, rag_response) collected_data["Current Stock Price"] = f"${current_stock_price:.2f}" if current_stock_price is not None else "N/A" answer = answer_question_with_data(f"{translation}", collected_data) return jsonify({"answer": answer}) else: with ThreadPoolExecutor() as executor: futures = { executor.submit(get_answer, user_input): "rag_response", executor.submit(summarizer.google_search, user_input): "google_results", executor.submit(summarizer.fetch_google_snippet, user_input): "google_snippet" } results = {futures[future]: future.result() for future in as_completed(futures)} google_results = results.get("google_results", "No additional news found through Google Search.") google_snippet = results.get("google_snippet", "Snippet not found.") rag_response = results.get("rag_response", "No response from RAG.") collected_data = prepare_data("", "", "", google_results, {}, google_snippet, rag_response) answer = answer_question_with_data(f"{user_input}", collected_data) return jsonify({"answer": answer}) except Exception as e: logger.error(f"An error occurred: {e}") return jsonify({"error": "An error occurred while processing your request. Please try again later."}), 500 if __name__ == '__main__': app.run(host='0.0.0.0', port=5000)