AarthiAI / README.md
PROSTLE's picture
fix: shorten HuggingFace short_description to under 60 chars
4359352
|
Raw History Blame Contribute Delete
5.08 kB
metadata
title: StockSense AI
emoji: πŸ“ˆ
colorFrom: indigo
colorTo: purple
sdk: docker
app_port: 8000
short_description: AI trading engine for Indian markets (NSE/BSE)

πŸ“ˆ StockSense-AI β€” Enterprise-Grade Trading Engine

StockSense-AI is a multi-factor hybrid intelligence platform that blends Deep Learning, Intraday ML, and LLM reasoning to provide high-precision stock signals for the Indian market (NSE/BSE).


🌟 Vision & Pitch

"StockSense AI doesn't just predict prices; it understands market context. By fusing LSTM-based daily forecasting with XGBoost intraday signals, FinBERT sentiment analysis, and Gemini LLM reasoning, we've built a decision engine that adapts to market regimes in real-time. Whether you're tracking 5-day trends or hunting for intraday breakouts, StockSense AI provides the confidence score and risk-mitigation strategy you need."


πŸš€ Key Features

🧠 The Hybrid Command Engine

  • LSTM Neural Network: 2-layer Deep Learning model trained on 2 years of daily data for 5-day price forecasting.
  • XGBoost Intraday Pipeline: High-velocity ML model trained on 1-minute bars to capture micro-trends.
  • FinBERT Sentiment Analysis: Advanced NLP scoring of real-time financial news and social sentiment.
  • Gemini LLM Reasoning: Human-readable trade justifications and "why" analysis for every signal.
  • Factor Blending: 5-factor weighted scoring (Technicals + Sentiment + LSTM + XGBoost + Market Context).

🏦 Paper Trading & Wallet Simulation

  • Virtual Portfolio Management: Track open positions, realized/unrealized P&L, and win rates.
  • Integrated Wallet: Deposit and withdraw (Razorpay-style simulation) with full transaction history.
  • Account Performance Charting: Visual snapshots of your portfolio growth over time.

πŸ›‘οΈ Advanced Risk Management

  • Dynamic ATR-based Levels: Stop-Loss and Take-Profit targets that adjust automatically to ticker volatility.
  • Adaptive Market Regimes: System switches behavior across 4 states: Bull Low-Vol, Bull High-Vol, Bear High-Vol, and Sideways.
  • Risk-Based Position Sizing: Automatic share calculation based on your "capital-at-risk" percentage.
  • Trailing Stop-Loss: Intelligent profit-locking mechanism that follows the price as it rises.

πŸ’Ž Glassmorphic Interactive UI

  • Premium Aesthetics: High-end translucent design system with vibrant neon accents.
  • Multi-Tab Dashboard: Switch between Charts, Unified Dashboard, Auto-Trade Engine, and Trading Portfolio.
  • Real-Time Price Banner: Live tracking of tickers with market-depth indicators.
  • High-Potential Scanner: Real-time scanner for "Strong Buy" signals across Top 50 NSE stocks.

πŸ—οΈ Architecture & Stack

graph LR
    User((User)) -- "Ticker" --> API[FastAPI Gateway]
    
    subgraph "Engine Layer"
        API --> LSTM[Deep Learning: Daily Trends]
        API --> XGB[XGBoost: Intraday Signals]
        API --> NLP[FinBERT: News Sentiment]
        API --> Gemini[LLM: Market Reasoning]
    end

    subgraph "Decision AI"
        LSTM & XGB & NLP & Gemini --> Blender[Weighted Composite Blender]
        Blender --> Regime[Regime/Volatility Filter]
    end

    subgraph "Execution Layer"
        Regime --> Risk[ATR Risk Engine: SL/TP/Sizing]
        Risk --> Sim[[Paper Trading Simulator]]
    end

Tech Stack

  • Backend: Python 3.10+, FastAPI, Uvicorn, SQLAlchemy (SQLite).
  • ML/Science: PyTorch (LSTM), XGBoost, Transformers (FinBERT), Scikit-learn, Pandas, yfinance.
  • Frontend: Vanilla HTML5/JS (ES6+), Chart.js for visualization, Glassmorphic CSS Engine.
  • Reasoning: Google Gemini 2.0 API.

πŸ”§ Installation & Setup

1. Backend (The Brain)

cd backend
python -m venv venv
# Windows
.\venv\Scripts\activate
# Linux/Mac
source venv/bin/activate

pip install -r requirements.txt
# Create a .env file with your GEMINI_API_KEY
uvicorn app:app --port 8000

2. Frontend (The Interface)

Simply host the frontend directory using any local server:

cd frontend
python -m http.server 5500

Visit http://localhost:5500.


πŸ“‹ ML Pipeline Details

Stage Logic
Data 1-min & Daily OHLCV via Yahoo Finance API.
LSTM 60-bar lookback window, recursive 5-day out-of-sample forecast.
XGBoost Feature-rich ensemble (RSI, MACD, BB, VWAP, Supertrend) for next-5-bar returns.
Regime 4-state classification using India VIX and Sector Strength.
Scoring Normal: 45% LSTM, 45% XGB, 10% Sent | Volatile: 48.5% LSTM, 48.5% XGB, 3% Sent.

⚠️ Disclaimer

StockSense AI is for educational and simulation purposes only. This is a paper-trading platform. Algorithms can fail, and market data can be delayed. Never use this tool for actual financial investments without consulting a certified advisor.


πŸ“œ License

Distributed under the MIT License. See LICENSE for more information.