Getting Started
How to use AI Engineering Patterns and find the right patterns for your system.
AI Engineering Patterns is a structured pattern library for engineers building AI systems in production. It is not an awesome list of links, not vendor documentation, and not a research paper collection. It is a curated set of named patterns with descriptions, trade-offs, when-to-use guides, and code examples for every layer of a production AI system.
How to Use This Resource
If you know what you are looking for
Browse the sidebar. Patterns are organized into 8 pillars. Each pillar covers a distinct layer of a production AI system.
If you are not sure where to start
Use the Decision Guide to answer a few questions about your system and get recommended patterns.
If you want to understand the landscape
Read through the pillars in order. They roughly follow the lifecycle of a request through a production AI system:
- Inference & Serving — How requests reach models and how responses come back.
- Data Patterns for AI — The data foundation every AI system depends on.
- Reliability & Resilience — Keeping AI systems working when things go wrong.
- Retrieval & Memory — What knowledge AI systems can access and remember.
- Observability & Monitoring — Seeing what your AI system is actually doing.
- Security & Trust — Guardrails, PII handling, and prompt injection defenses.
- Cost & Efficiency — Running AI systems without unbounded spend.
- Governance & Compliance — Lineage, model cards, and compliance operations.
- Graph Patterns — Knowledge graphs, graph-based reasoning, and entity resolution.
- Evaluation & Testing — LLM-as-Judge, eval pipelines, and regression testing.
Reading a Pattern
Every pattern follows the same structure:
| Section | Purpose |
|---|---|
| What It Is | One paragraph plain-language description |
| The Problem It Solves | What breaks without this pattern |
| How It Works | Step-by-step mechanism with diagrams |
| When to Use It | Specific conditions and workload types |
| When not to Use It | Explicit anti-use-cases |
| Trade-offs | 2-4 honest trade-offs |
| Failure Modes | What breaks when the pattern is deployed correctly — trigger, symptom, mitigation |
| Implementation Example | Minimal working code snippet |
| Tool Landscape | Tools that implement or support the pattern |
| Related Patterns | Links to adjacent patterns |
| Further Reading | 2-4 high-quality external references |
The When not to Use It section is the most important. It is what makes this resource opinionated rather than encyclopedic.
Pattern Maturity
Each pattern has a status indicating its maturity level:
- Proposed — Pattern identified, not yet validated in production systems.
- Emerging — Used by early adopters, trade-offs still being understood.
- Validated in Production — Widely used by multiple teams, trade-offs well understood.
Contributing
This is an open-source project. If you have used a pattern in production and want to share it, or if you see something that needs correction, see the Contributing Guide.