Skip to content

Systems

The architecture components I reuse for production AI systems — agents, memory, orchestration, retrieval, and tools — for engineers and technical buyers.

AI Systems EngineeringAgents connect to memory and retrieval. Runtime, tools, evaluations, and orchestration form the operating layer around AI systems engineering.AI SystemsengineeringAgentsMemoryRetrievalRuntimeToolsEvalsOrchestration
  • agents to memory: primary
  • agents to retrieval: primary
  • memory to runtime: secondary
  • retrieval to tools: secondary
  • runtime to evals: feedback
  • tools to orchestration: feedback
  • evals to orchestration: primary
  • runtime to tools: secondary

Recurring components

The architecture patterns I reuse across projects.

Multi-Agent Runtime

Coordination, state transitions, and typed execution boundaries for agents that need to work together.

#orchestration#delegation#coordination#state persistence

Memory Layer

Episodic memory, semantic retrieval, compression, and replayable context for agents that need continuity.

#episodic memory#semantic retrieval#compression#persistence

Workflow Engine

Durable execution, retries, branching, checkpoints, and human gates for AI workflows in production.

#durable execution#retries#branching#human-in-the-loop

Retrieval Infrastructure

Embedding pipelines, hybrid search, metadata filters, reranking, and stable eval sets for RAG systems.

#embeddings#reranking#hybrid search#vector pipelines

Tool Ecosystem

MCP integrations, typed tools, permission boundaries, routing schemas, and protocol adapters.

#MCP integrations#typed tools#routing#protocol adapters

Evaluation Systems

Regression suites, adversarial cases, trace review, cost tracking, and evidence that a system is ready to ship.

#evals#regression#observability#failure recovery

Technical pages

One hub per topic: each connects the concept to its components, projects, and articles.

Multi-Agent Systems

Architecture patterns for coordinating multiple AI agents with clear roles, shared state, tools, and failure handling.

Agent Orchestration

Routing, delegation, checkpoints, and supervision patterns for agent workflows in production.

AI Workflow Engines

Durable workflow design for AI systems that need branching, retries, persistence, and audit trails.

Durable AI Execution

Patterns for AI workflows that survive partial failure and remain inspectable after each run.

Memory Architectures

Memory layers for AI systems: semantic recall, episodic traces, compression, persistence, and retrieval evaluation.

MCP Ecosystem

Model Context Protocol integrations, connector identity, tool boundaries, and operational interfaces for agents.

Retrieval Infrastructure

Hybrid search, embedding pipelines, metadata strategy, reranking, and evals for RAG systems in production.