Trading AI Orchestrator

class TradingAIOrchestrator:
    def __init__(self):
        self.gemini = GeminiClient(api_key=GEMINI_KEY)
        self.deepseek = DeepSeekClient(api_key=DEEPSEEK_KEY)
        self.llama = OllamaClient(model="llama3.3:70b")
        self.consensus_threshold = 0.65
        
    async def analyze_swing(self, symbol: str, data: dict):
        """Core de Swing Trading - Gemini lidera"""
        # Gemini: Análisis macro y sentimiento
        macro = await self.gemini.analyze(
            prompt=f"Análisis macro y sentimiento para {symbol}",
            context=data['news'] + data['fundamentals'],
            model="gemini-3.7-flash"
        )
        # DeepSeek: Confirmación técnica de swing
        technical = await self.deepseek.analyze(
            prompt=f"Patrones de swing en {symbol}",
            context=data['candles_h4'],
            reasoning=True  # Chain-of-Thought
        )
        # LLaMA: Timing de entrada optimizado
        timing = await self.llama.analyze(
            prompt=f"Microestructura para entrada swing",
            context=data['orderbook'],
            temperature=0.1  # Determinista
        )
        return self.consensus_vote([macro, technical, timing])
    
    async def analyze_scalping(self, symbol: str, data: dict):
        """Core de Scalping - LLaMA lidera"""
        # LLaMA: Análisis ultra-rápido de microestructura
        micro = await self.llama.analyze(
            prompt=f"Order book imbalance y FVG para {symbol}",
            context=data['ticks_m1'],
            max_tokens=150  # Respuesta rápida
        )
        # DeepSeek: Confirmación de momentum
        momentum = await self.deepseek.analyze(
            prompt=f"Momentum confirmation scalping",
            context=data['indicators_m5'],
            reasoning=True
        )
        # Gemini: Filtro de noticias de último momento
        news_filter = await self.gemini.analyze(
            prompt=f"Breaking news impact assessment",
            context=data['headlines'],
            model="gemini-3.1-flash-lite"  # Rápido
        )
        return self.consensus_vote([micro, momentum, news_filter])
    
    def consensus_vote(self, signals: list) -> dict:
        """Motor de consenso ponderado"""
        weights = {
            'swing': {'gemini': 0.5, 'deepseek': 0.3, 'llama': 0.2},
            'intraday': {'gemini': 0.25, 'deepseek': 0.45, 'llama': 0.3},
            'scalping': {'gemini': 0.15, 'deepseek': 0.25, 'llama': 0.6}
        }
        # ... lógica de votación y decisión final
                        

# Configuración del Core Multi-IA
ai_orchestrator:
  
  # Google Gemini - Análisis Macro & Sentimiento
  gemini:
    api_key: "${GEMINI_API_KEY}"
    model_primary: "gemini-3.7-flash"
    model_fallback: "gemini-3.1-flash-lite"
    temperature: 0.3
    max_tokens: 2048
    role: "macro_sentiment_analyst"
    weight_swing: 0.50
    weight_intraday: 0.25
    weight_scalping: 0.15
    
  # DeepSeek - Razonamiento Técnico Avanzado
  deepseek:
    api_key: "${DEEPSEEK_API_KEY}"
    model: "deepseek-reasoner"
    temperature: 0.2
    max_tokens: 4096
    reasoning_effort: "high"
    role: "technical_pattern_analyst"
    weight_swing: 0.30
    weight_intraday: 0.45
    weight_scalping: 0.25
    
  # LLaMA 3 (Local vía Ollama) - Microestructura
  llama:
    base_url: "http://localhost:11434"
    model: "llama3.3:70b-instruct-q4_K_M"
    temperature: 0.1
    max_tokens: 512
    role: "microstructure_scalping"
    weight_swing: 0.20
    weight_intraday: 0.30
    weight_scalping: 0.60

# Estrategias por Timeframe
strategies:
  swing:
    timeframe: "H4"
    ai_leader: "gemini"
    consensus_mode: "weighted"
    min_confidence: 0.75
    holding_period: "2-10 días"
    
  intraday:
    timeframe: "M15"
    ai_leader: "deepseek"
    consensus_mode: "majority"
    min_confidence: 0.70
    holding_period: "1-8 horas"
    
  scalping:
    timeframe: "M1"
    ai_leader: "llama"
    consensus_mode: "unanimous"
    min_confidence: 0.85
    holding_period: "5-30 minutos"