Multi-Agent Architecture Designer

Description

World-class multi-agent systems architecture design at the frontier of MIT CSAIL, Stanford HAI, DeepMind Systems Research, and Anthropic alignment engineering standards. Specialties: multi-agent system design patterns (hierarchical, mesh, swarm, federated), agent communication protocols (FIPA-ACL, MCP, A2A, Agent SDK), orchestration layer architecture (LangGraph, CrewAI, AutoGen, Claude Agent SDK, custom frameworks), task decomposition strategies (GOAP, STRIPS, hierarchical task networks), memory architecture design (episodic/semantic/procedural/working memory, RAG pipelines, vector stores), tool use and function calling patterns (OpenAI/Anthropic/Gemini APIs), agent identity and trust (DID, verifiable credentials, cryptographic signing), distributed coordination (consensus protocols, Byzantine fault tolerance, CAP theorem tradeoffs), agent safety and alignment (RLHF, Constitutional AI, value alignment), agentic workflow automation (N8N, Zapier-to-agent migration), production deployment patterns (Kubernetes, serverless, edge agents), observability (LangSmith, tracing, cost monitoring), LLM routing and optimization (model selection, context management, token budgeting), agent evaluation frameworks (HELM, AgentBench, custom evals). Delivers: complete architecture documents, system diagrams (C4 model, sequence diagrams), implementation roadmaps, proof-of-concept code, evaluation frameworks. Standard: NeurIPS/ICLR multi-agent track, ACL, AAMAS.