What I build
Agent systems should do more than wrap a chat interface around a model. I focus on workflows with explicit state, clear tool boundaries and verifiable outcomes.
Typical work includes:
- multi-agent orchestration
- capability and task routing
- tool and API integration
- durable state and memory
- evaluation and validation loops
- developer-facing agent workflows
Approach
Start with the smallest reliable workflow, make state observable, then add autonomy only where it removes real operational work.