MarketSage is a sophisticated, AI-driven application designed to forecast stock prices and deliver comprehensive financial analysis. By integrating a Long Short-Term Memory (LSTM) model to capture sequential patterns with a Transformer network's attention mechanisms, this hybrid system provides robust, data-backed insights. The entire application is built for local execution using Streamlit, ensuring a polished, intuitive, and privacy-focused experience.

• Hybrid AI Model: Combines the strengths of LSTM and Transformer architectures for high-fidelity price predictions.
• Comprehensive Analysis: Generates a detailed report complete with key insights, risk factors, and essential technical indicators (RSI, MACD, EMAs).
• AI-Powered Insights: Integrates with the Google Gemini API to produce a narrative, professional-grade financial analysis that contextualizes market data.
• Intuitive UI: A clean, interactive Streamlit dashboard lets users easily select a stock, view historical data, and visualize future predictions.
• Local & Private: All models are trained and run on your local machine, ensuring complete data privacy and offline functionality.

MarketSage/
├── src/
│   ├── app.py          # Main Streamlit application dashboard
│   ├── main.py         # Core model training script
│   └── marketsage.py   # NEW: Interactive setup & run script
├── prompts/
│   └── gemini.txt      # Template for customizing AI-generated insights
├── notebooks/
│   └── ...             # Jupyter notebooks for model prototyping
├── saved_models/
│   └── ...             # Directory for all trained models & data
├── .env.example        # Example for environment variables
└── README.md

A comprehensive report is available in the documentation/MarketSage.md file. It covers:

• Problem definition and motivation
• Related work in stock market prediction
• Detailed methodology (data pipeline, model architectures, hybrid ensemble)
• Experiments, results, and visualizations
• Future enhancements and references

Open MarketSage.md

You can tailor the AI-generated reports by modifying the prompts/gemini.txt file. This template is dynamically filled with real-time stock data.

How to Use:

1. Open prompts/gemini.txt and edit the content.
2. Use placeholders in curly braces, like {company_name} or {current_price}, to pull in dynamic data.
3. Supported variables include: company_name, symbol, current_price, sector, model_accuracy, predictions_1_week_gain_pct, predictions_1_month_gain_pct, RSI, MACD, MACD_Signal, EMA20, EMA50, ATR, and recommendation_score.

Follow these steps to set up and run MarketSage on your local machine.

Open your terminal and clone the repository:

git clone https://github.com/lazy-coder-03/MarketSage.git
cd MarketSage

The marketsage.py script simplifies the entire workflow. It handles virtual environment creation, dependency installation, model training, and application launch with a single command.

python src/marketsage.py

This script will guide you by asking for a stock symbol (e.g., RELIANCE.NS, TCS), handling various input formats. If a model for that stock already exists, it will prompt you to either use the existing one or retrain it. Once the model is ready, the Streamlit app will automatically launch in your browser.

To enable the AI-Powered Insights feature, you will need a Google Gemini API key.

1. Create a copy of .env.example and rename it to .env.
2. Get your API key from Google AI Studio.
3. Add your key to the .env file:

GEMINI_API_KEY="YOUR_API_KEY_HERE"

This project is for educational and research purposes only. Feel free to fork the repository, modify the models, and contribute to its development.

Below are sample screenshots of MarketSage's UI and features:

Feature	Screenshot
Welcome Page
Sidebar & Model Selection
Historical Price Chart
Stock Performance (52W Range)
Current Market Status
Technical Indicators Table
AI Recommendation
AI Predictions
Analysis Chart with Predictions
Gemini AI Insights
Model Performance Metrics