Production AI Patterns
14 named patterns across 10 pillars. Trade-offs, implementation guides, and code examples for every layer of a production AI system.
Three ways into the library
Section titled “Three ways into the library” Browse the catalog Scan the full library when you already know the shape of the system you need.
Use the decision guide Answer a few questions and narrow the pattern set before you build.
See the pattern graph Explore adjacent ideas and dependencies across pillars.
Pick a pillar
Section titled “Pick a pillar” Inference & Serving Gateway routing, semantic caching, model routers, and fallback chains.
Data Patterns Data contracts, feature stores, training pipelines, and eval datasets.
Reliability & Resilience Circuit breakers, graceful degradation, canary deployments, and model rollback.
Retrieval & Memory Hybrid search, RAG, reranking, and contextual memory.
Observability Span-level tracing, cost attribution, and quality drift detection.
Security & Trust Input sanitization, output validation, audit trails, and PII scrubbing.
Cost & Efficiency Token budgets, tiered models, prompt compression, and cost circuit breakers.
Governance & Compliance Model cards, data lineage, policy-as-code, and model versioning.
Graph Patterns GraphRAG, graph-based reasoning, and entity resolution.
Evaluation & Testing LLM-as-Judge, eval pipelines, regression testing, and benchmarking.
Why this exists
Section titled “Why this exists” Not an awesome list Structured, opinionated patterns with named trade-offs and explicit anti-use-cases.
Not vendor docs Framework-agnostic, model-agnostic, vendor-neutral. The patterns, not the products.
Contribute a pattern Open-source and community-maintained. Every pattern goes through structured review.