Multi-Agent Systems
Designing systems where several LLM agents delegate, hand off and cooperate — a core focus of my open-source work.
A multi-agent system splits work across several LLM agents — a supervisor and workers, subagents called as tools, or peers that hand control to each other — so each agent keeps a focused context and toolset.
How I use it
- Akashi — multi-agent orchestration is one of its three pillars: subagent-as-tool with isolated context, control handoffs (
transfer_to_<name>), per-step model escalation, and theakashi_workflowpackage for deterministic, bounded-concurrency fan-out. - langgraph-ts-agents — a supervisor routes between researcher, analyst and writer workers, with a Human-in-the-Loop gate before the writer.
- langgraph-langserve-starter — a planner → researcher → synthesizer graph served over HTTP.
- Somnio AI Tools — I refactored the CLI to a data-driven registry of 17 AI coding agents.
- Conversational AI Agents in my Independent AI & Full-Stack Engineer (Freelance) work, and Somnio Software, where multi-agent patterns are part of the stack.
Credentials and courses
Multi-Agent Systems with LangGraph in the Agentic AI Engineering Specialization specialization.