Real-time Signal Logic

Real-Time Signal Logic

Signal.Engine utilizes a "Hybrid Brain" architecture to generate trading decisions. It doesn't just predict price movement; it synthesizes technical indicators, sequence modeling (LSTM), and reinforcement learning (PPO) to produce actionable signals with human-readable rationale.

The Signal Generation Flow

The system operates on a continuous "Think" loop. This process can be triggered manually via the dashboard or automated through API polling.

1. Data Ingestion: The system fetches the latest market candles for the defined universe (e.g., Nifty 500 or S&P 500).
2. Inference:
   • LSTM Layer: Analyzes the last 50 candles to identify time-series patterns.
   • Heuristic Layer: Evaluates volatility, RSI, and MACD.
3. Synthesis: The HybridBrain merges these inputs to produce a Decision Object.
4. Risk Overlay: The Simulation Engine calculates Quantitative Risk (WinRate, VaR, EV) before presenting the signal.

API Reference: Consuming Signals

The backend exposes several endpoints to interact with the signal logic.

1. Triggering a Scan

GET /api/scan

Starts the HybridBrain.think() process in a background thread. This prevents the API from locking up during complex calculations.

• Response:

  {
    "status": "started",
    "message": "Scan triggered in background."
  }
  

2. Fetching Results

GET /api/results

Retrieves the most recent decisions generated by the brain, along with the current simulation state.

• Response Schema: | Field | Type | Description | | :--- | :--- | :--- | | data | Array<Decision> | List of active stock signals. | | simulation | Object | Current portfolio balance, equity, and PnL. | | is_thinking | Boolean | Whether a background scan is currently running. | | logs | Array<String> | Recent processing steps and system events. |

The Decision Object

Every signal generated by the system follows a strict schema designed for both automated execution and user review.

{
  "Ticker": "RELIANCE.NS",
  "Action": "BUY",
  "Confidence": 0.89,
  "Rational": [
    "LSTM Sequence shows bullish reversal",
    "RSI Oversold (28.4)",
    "High Volatility detected"
  ],
  "QuantRisk": {
    "WinRate": 0.65,
    "EV": 1.25,
    "VaR95": -2.4
  }
}

Key Fields:

• Action: Can be BUY, SELL, or HOLD (Neutral).
• Confidence: A value from 0.0 to 1.0.
• Rational: An array of strings explaining why the brain made this choice. This moves the system from a "Black Box" to "Explainable AI."
• QuantRisk: Metrics providing the Expected Value (EV) and Value at Risk (VaR) for the specific trade.

Real-Time Alerts & UI Integration

The frontend implementation (App.tsx) includes logic for real-time monitoring and alerting based on signal strength.

High-Confidence Notifications

The system includes a NotificationService that triggers browser alerts when the brain's confidence exceeds a specific threshold:

• Threshold: Confidence >= 0.85
• Behavior: High-confidence signals trigger a push notification. Signals with Confidence >= 0.90 require user interaction to dismiss, highlighting high-probability setups.

Automatic Polling Loop

When the dashboard is set to Auto Mode, the frontend polls the /api/scan and /api/results endpoints every 2 seconds. This ensures the "Market Mood" and signal list stay synchronized with the latest exchange data.

// Example of the frontend polling logic
useEffect(() => {
    if (isAuto) {
        interval = setInterval(() => {
            if (!loading) runScan(true); // silent background update
        }, 2000);
    }
    return () => clearInterval(interval);
}, [isAuto]);

Market Regime Detection

Beyond individual stock signals, the logic derives a global Market Mood:

• VOLATILE: Triggered if >30% of signals report "High Volatility" in their rationale.
• CALM: Triggered when price action stabilizes across the scanned universe.

This regime detection influences the PPO Agent's risk appetite, allowing for tighter stop-losses during volatile periods.