Institutional-Grade Intraday Quant System (NSE)

An institutional-grade, fully autonomous quantitative trading framework designed for the Indian Stock Market (NSE). Unlike basic retail bots that rely on simple moving averages, this system implements advanced hedge-fund methodologies including Meta-Labeling, Purged Walk-Forward Cross Validation, VWAP Execution Algorithms, and a Headless Broker Auto-Login.

---

🚀 Key Features

• Advanced Machine Learning Ensemble: Utilizes a primary ensemble of LightGBM and XGBoost classifiers trained on high-fidelity features (Order Flow Imbalance, VPIN, Micro-price Volatility) to detect intraday alpha.
• CatBoost Meta-Labeler & Conformal Prediction: Instead of blindly trusting the primary model, a secondary CatBoost meta-labeler predicts the probability of success of the primary signal. It uses conformal prediction bands to dynamically gate trades based on market regime and uncertainty, effectively filtering out 99% of market noise.
• Algorithmic Execution (OMS): Features a robust Order Management System (OMS) utilizing a VWAP (Volume-Weighted Average Price) algorithm. Large institutional-sized orders are intelligently sliced into child orders based on historical intraday U-shaped volume curves to minimize market impact and slippage.
• Headless Broker Integration: Fully automated fyers-apiv3 integration. The system programmatically generates its own TOTP codes (pyotp) and bypasses the manual daily OAuth web-login flow, allowing the daemon to run 100% autonomously.
• Dynamic Pre-Market Screener: Uses highly parallelized data fetching to scan the entire F&O universe at 09:15 AM, automatically identifying the top "Stocks in Play" based on pre-market gap momentum and relative volume.
• Institutional Risk Management: Hardcoded Risk:Reward gating, India VIX volatility filters, and API circuit breakers ensure capital preservation in destructive market conditions.

---

🧠 System Architecture

graph TD
    A[Dynamic Screener] -->|Top Momentum Stocks| B(Market Data Engine)
    B -->|OHLCV & Tick Data| C{Feature Store}
    
    subgraph Machine Learning Pipeline
    C -->|Micro-structure Features| D[Primary Ensemble]
    D -.->|Base Signal| E(CatBoost Meta-Labeler)
    E -->|Confidence Score| F[Call Generator]
    end
    
    F -->|Risk:Reward Filter| G[Order Management System]
    
    subgraph Execution
    G -->|Large Order| H(VWAP Slicer)
    H -->|Child Orders| I[Fyers Broker Client]
    I -->|Live Market| J((NSE))
    end

Loading

---

🛠️ Installation

Prerequisites

• Python 3.10+
• Microsoft Visual C++ Build Tools (Required for aiohttp on Windows)
• An active Fyers Trading Account (for live execution)

Setup

1. Clone the repository:

   git clone https://github.com/yourusername/intraday-quant-system.git
   cd intraday-quant-system

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # On Windows use: .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Configure Environment Variables: Create a .env file in the root directory and add your Fyers credentials:

   FYERS_APP_ID="YOUR_APP_ID-100"
   FYERS_SECRET_KEY="YOUR_SECRET_KEY"
   FYERS_CLIENT_ID="YOUR_FY_ID"
   FYERS_TOTP_SECRET="YOUR_32_CHAR_BASE32_SECRET"
   FYERS_PIN="1234"

---

💻 Usage

1. Live Trading Daemon (Production)

Run the headless live trading daemon. This will automatically authenticate, run the dynamic screener, and enter the live execution tick loop:

python -m scripts.live_trader --top-n 10

2. Manual Call Generation (Testing)

Run the core ML pipeline to generate trade calls for the top 5 momentum stocks using the yfinance fallback engine (useful for testing without live broker execution):

python -m scripts.generate_calls --dynamic-screener --top-n 5

---

⚠️ Disclaimer

This software is for educational and research purposes only. Do not risk money which you are afraid to lose. USE AS AT YOUR OWN RISK. The authors assume no responsibility for your trading results.