Routing Architecture

# Hybrid Routing Decision Flow User Query: "analyze my forehand consistency" | v +---------------------------+ | Rule-Based Router | | Pattern: "forehand" | | Match: TennisAgent | | Confidence: 0.85 | +---------------------------+ | | confidence >= 0.5? | +---> YES: Route to TennisAgent | +---> NO: LLM Fallback | v +------------------+ | LLM Router | | "Which domain | | handles this?" | +------------------+ | v Route to selected domain

# RoutingResult dataclass @dataclass class RoutingResult: domain: str # "TennisAgent" confidence: float # 0.85 method: str # "rule" | "llm" | "hybrid" matched_patterns: List[str] # ["forehand", "tennis"] reasoning: Optional[str] # LLM explanation (if used) # Example Result RoutingResult( domain="TennisAgent", confidence=0.85, method="rule", matched_patterns=["forehand", "consistency"], reasoning=None )

# LLM Router System Prompt You are a query router for Project Phoenix domains. Given a user query, determine which domain should handle it. Available domains: - TennisAgent: Tennis sensor analysis, match/practice data - Portfolio: Investment portfolio, tax, options - Optiver: Trading ML, Kaggle competition - WQ: WorldQuant data science projects - Stan: Stanford Stats 315A coursework - AI_WQ: Computer vision, deep learning - QCiAgent: Quantum computing products - YieldModel: Semiconductor manufacturing - ParableAgent: Biblical parables analysis - PPR_Agent: Medical device reports Respond with JSON: {"domain": "...", "confidence": 0.X, "reasoning": "..."}