多模型智能路由决策引擎提示词

You are an AI model routing architect. Design a complete intelligent model routing system based on my requirements.

Task

Create a routing decision engine that automatically selects the optimal LLM for each request based on:

• Task complexity (simple Q&A vs. multi-step reasoning vs. code generation)
• Quality requirements (draft vs. production vs. critical)
• Cost budget (per-request and monthly caps)
• Latency requirements (real-time vs. batch)
• Context length needs

Output Requirements

1. Task Classification Rules

Define a taxonomy of request types with example patterns:

• Tier 1 (cheap/fast model): Simple lookups, formatting, translations
• Tier 2 (balanced model): Summarization, basic analysis, standard code
• Tier 3 (premium model): Complex reasoning, architecture design, novel code
• Tier 4 (frontier model): Research, multi-step planning, critical decisions

2. Routing Logic (Pseudocode)

Provide implementable routing logic with:

• Pre-routing classifier prompt (< 100 tokens)
• Fallback/escalation rules (auto-retry with stronger model on failure)
• Cost tracking and budget enforcement
• A/B testing hooks for continuous optimization

3. Model Portfolio

Recommend specific models for each tier with:

• Cost per 1M tokens (input/output)
• Strengths and weaknesses
• When to prefer each

4. Monitoring Dashboard Metrics

• Cost per task category
• Quality score distribution by model
• Routing accuracy (was the right model chosen?)
• Monthly cost projection

My current setup: [Describe your models, use cases, and monthly budget]

Provide the complete routing system design with ready-to-implement configurations.