Ai Patterns

Routing & Optimization

7. Router Pattern — Classify intent, route to appropriate model/tool
8. Skills / Lazy Loading — Load tool definitions on-demand (critical with 10+ tools)
9. Model Selection — Dense models for complex reasoning, sparse/MoE for high-volume

Caching & Performance

10. Semantic Caching — Return cached response for semantically similar queries. Requires tenant isolation.
11. LLM Gateway — Centralized proxy for auth, rate limiting, failover across providers

Security & Safety

12. Sanitization Middleware — Input/output filtering. Never expose a raw LLM directly.
13. Prompt Injection Defense — Validate, sandbox, detect malicious instructions
14. PII Protection — Detect and redact PII, scope caches to users

Architecture Patterns

15. RAG — Augment LLM with external knowledge retrieval
16. Multi-Agent Orchestration — Sequential, parallel, hierarchical, or debate patterns
17. Flow Engineering — Design multi-step workflows for complex tasks

Implementation Checklist

• Input handling: structured inputs or sanitized templates?
• Output validation: structured outputs or post-processing?
• Caching strategy: semantic caching with tenant isolation?
• Routing: appropriate model selection per query type?
• Security: guardrails for input/output? Prompt injection defense?
• Resilience: LLM gateway or fallback strategy?
• Cost optimization: context caching? Router pattern?
• Observability: logging, metrics, evaluation loops?