Agent Orchestration
Coordinating LLM agents, tools and steps into reliable workflows — the hub concept behind most of my AI engineering.
Agent orchestration is the control layer around LLM agents: deciding which agent or tool runs next, passing state between steps, running work in parallel, retrying failures and pausing for people. It can be model-driven (the LLM chooses the next step) or code-driven (a deterministic workflow decides).
How I use it
- Akashi — both styles: model-driven subagent-as-tool, handoffs and model escalation, and the code-driven
akashi_workflowpackage with bounded-concurrency fan-out, typed pipelines, retries with backoff, timeouts and cancellation. - langgraph-ts-agents and langgraph-langserve-starter — supervisor/worker graphs in LangGraph.
- Somnio AI Tools — I built a custom workflow system for multi-step, multi-model AI task orchestration in the open-source CLI.
- Conversational AI Agents — orchestrated agents behind a gateway in my Independent AI & Full-Stack Engineer (Freelance) work.