Skip to main content

​
Overview

The AI Decision Engine is the brain of the XAUUSD Trading Assistant, powered by Groq’s large language model (LLM) and integrated through the LangChain framework. This component transforms raw technical analysis into actionable trading signals with contextual understanding.
Groq provides ultra-fast LLM inference, enabling real-time analysis of complex market data without significant latency.

​
Architecture

​
Component Stack

┌──────────────────────────────────┐
│     Technical Features Input     │
│  (Multi-timeframe indicators)    │
└────────────┬─────────────────────┘
             │
             ▼
┌──────────────────────────────────┐
│         LangChain Layer          │
│  - Prompt Engineering            │
│  - Context Management            │
│  - Response Parsing              │
└────────────┬─────────────────────┘
             │
             ▼
┌──────────────────────────────────┐
│          Groq LLM API            │
│  - Pattern Recognition           │
│  - Market Analysis               │
│  - Signal Generation             │
└────────────┬─────────────────────┘
             │
             ▼
┌──────────────────────────────────┐
│       Trading Signal Output      │
│  - Direction (Buy/Sell/Hold)     │
│  - Entry Price                   │
│  - Stop Loss                     │
│  - Take Profit                   │
│  - Reasoning                     │
└──────────────────────────────────┘

​
How It Works

​
1. Feature Extraction

The XAUUSDTradingBot analyzes market data and extracts comprehensive technical features:

Trend Indicators

  • EMA positions across all timeframes
  • Price action relative to moving averages
  • Trend strength and direction

Momentum Indicators

  • RSI values for each timeframe
  • Overbought/oversold conditions
  • Momentum divergences

Volatility Metrics

  • ATR values for risk assessment
  • Market volatility state
  • Dynamic range calculations

Market Structure

  • Order blocks identification
  • Fair Value Gaps (FVG)
  • Support and demand zones

​
2. Prompt Engineering

LangChain enables sophisticated prompt construction that provides the LLM with structured context about market conditions, trading rules, and desired output format.
The system constructs prompts that include:
  • Market Context: Current timeframe analysis across D1, H4, H1, M30, M15, M5
  • Technical Features: All calculated indicators with their values
  • Trading Rules: Risk parameters, position sizing, and strategy guidelines
  • Output Format: Structured format for signals with entry, SL, and TP levels
Example prompt structure: 
Analyze the following XAUUSD market data across multiple timeframes:

Daily (D1): [RSI, EMA, ATR, Order Blocks]
H4: [RSI, EMA, ATR, FVG]
H1: [RSI, EMA, ATR]
...

Provide a trading signal with:
- Direction (Buy/Sell/Hold)
- Entry price
- Stop-loss (based on ATR)
- Take-profit targets
- Confidence level
- Reasoning

​
3. LLM Processing

Groq’s LLM processes the technical features and: Analyzes Patterns
  • Recognizes complex chart patterns across timeframes
  • Identifies confluence zones where multiple indicators align
  • Detects divergences that signal potential reversals
Evaluates Confluence
  • Assesses agreement between different timeframes
  • Weighs higher timeframe signals more heavily
  • Determines overall market bias
Generates Signals
  • Creates actionable buy/sell/hold recommendations
  • Calculates optimal entry points
  • Sets risk-appropriate stop-losses and take-profits
  • Provides detailed reasoning for transparency

​
4. Response Parsing

The system formats LLM responses for easy consumption in the Streamlit dashboard, ensuring traders get clear, actionable information.
LangChain handles:
  • Extracting structured data from LLM responses
  • Formatting signals for dashboard display
  • Error handling and validation
  • Maintaining response consistency

​
Integration with Trading Bot

The AI engine is invoked through the run_analysis() method: 
# From app.py
bot = XAUUSDTradingBot(api_key=st.secrets["GROQ_API_KEY"])
result = bot.run_analysis()

# Returns:
{
    'technical_features': '...',  # AI analysis summary
    'trading_signal': '...',      # Actionable signal
    'market_data': {...},         # Multi-timeframe data
    'current_spread': 'X points'  # Current spread
}

​
Key Advantages of AI-Powered Analysis

Pattern Recognition

LLMs excel at recognizing complex patterns that might be missed by traditional rule-based systems, identifying subtle market structures across multiple timeframes.

Contextual Understanding

The AI understands market context beyond raw numbers, considering factors like trend strength, volatility regime, and multi-timeframe alignment.

Adaptive Learning

While the current model doesn’t retrain in real-time, Groq’s LLM has been trained on vast amounts of financial data, providing robust market understanding.

Explainable Signals

Unlike black-box algorithms, the LLM provides reasoning for each signal, helping traders understand the “why” behind recommendations.

​
API Configuration

The Groq API is configured through Streamlit secrets: 
# .streamlit/secrets.toml
GROQ_API_KEY = "your-groq-api-key"
Keep your API key secure and never commit it to version control. Use environment variables or Streamlit secrets for production deployments.
The key is accessed in the application: 
bot = XAUUSDTradingBot(api_key=st.secrets["GROQ_API_KEY"])

​
Performance Characteristics

Speed
  • Groq’s LPU (Language Processing Unit) architecture provides near-instant inference
  • Typical analysis completion in under 2 seconds
  • Enables real-time decision making
Accuracy
  • Achieved 65% profitability rate in live trading tests
  • Higher accuracy when multiple timeframes show strong confluence
  • Conservative signal generation reduces false positives
Reliability
  • Comprehensive error handling in the bot implementation
  • Graceful degradation if API is unavailable
  • Dashboard displays clear error messages when issues occur

​
Signal Types Generated

The AI generates three types of signals:

​
Buy Signal

Direction: BUY
Entry: 2045.50
Stop Loss: 2042.30 (based on ATR)
Take Profit 1: 2050.00
Take Profit 2: 2055.00
Confidence: High
Reasoning: Daily and H4 uptrend confirmed, 
H1 RSI rebounding from oversold, bullish 
order block on M30...

​
Sell Signal

Direction: SELL
Entry: 2045.50
Stop Loss: 2048.70 (based on ATR)
Take Profit 1: 2040.00
Take Profit 2: 2035.00
Confidence: Medium
Reasoning: Daily downtrend, H4 bearish FVG, 
H1 RSI overbought...

​
Hold Signal

Direction: HOLD
Reasoning: Conflicting signals across timeframes. 
Daily shows uptrend but H4 and H1 showing 
bearish divergence. Wait for clarity...

​
Dashboard Integration

The AI-generated signals are displayed in the Trading Signal tab:
The dashboard separates technical analysis from trading signals, allowing traders to review the underlying data before acting on AI recommendations.
  • Tab 1 - Analysis: Raw technical features and multi-timeframe data
  • Tab 2 - Trading Signal: AI-generated actionable signal with reasoning

​
Future Enhancements

Potential improvements to the AI decision engine:
  • Integration with additional LLM providers for redundancy
  • Fine-tuning on historical XAUUSD data for better accuracy
  • Sentiment analysis integration from news sources
  • Multi-model ensemble for higher confidence
  • Reinforcement learning for strategy optimization
The modular architecture allows for easy integration of additional AI capabilities without disrupting the existing system.

Build docs developers (and LLMs) love

Get started for freeTalk to us