LLM Advisor
An autonomous trading agent that combines statistical mean reversion with LLM-based sentiment analysis for risk management.
Key Performance Indicators
Agentic Workflow
1. Sentiment Analysis
Every 15 minutes, the system feeds headlines and market context into Gemini 1.5 Flash. The LLM outputs a market-state score and risk multipliers used by execution guards.
2. Statistical Execution
The core engine calculates Z-scores on price action. If the Z-score exceeds the dynamically adjusted threshold, it executes mean reversion trades via Alpaca.
Live Monitoring Dashboard
This section is now backed by a real telemetry API using Supabase storage with local artifact fallback.
N/A
N/A
N/A (7d)
1d: N/A | 30d: N/A
0
Count in last 7 days of sample window.
N/A
0 trades available.
N/A
Avg win: N/A | Avg loss: N/A
NO FEED
Loop: N/A | Symbols: N/A
Recent Trades
No trades found in telemetry yet.
Model Visuals
Feature-importance plots from current training artifacts.
SPY feature importance
QQQ feature importance
IWM feature importance
Automated Risk Manager
Safety logic includes hard-coded circuit breakers that override AI decisions at configured drawdown thresholds.
def check_risk_parameters(current_pnl, max_drawdown_limit):
"""
Hard stop if we exceed daily loss limit.
"""
if current_pnl < -max_drawdown_limit:
logger.critical(f"Daily stop loss hit: {current_pnl}")
return {
"can_trade": False,
"action": "LIQUIDATE_ALL",
"reason": "MAX_DRAWDOWN_HIT"
}
# ... other checks (exposure, volatility) ...
return {"can_trade": True}Data Caveat
Current metrics are sourced from backtest artifacts and live-loop telemetry. Once real broker fills are persisted into the same schema, the dashboard will reflect true production P/L instead of backtest-only trade outcomes.