Project: Agent Workflow Orchestrator
- Difficulty: Advanced
- Primary stack: TypeScript or Python workflow runtime
- Estimated duration: 2 weeks
- Primary hiring signal: orchestration and state management
- Primary monetization signal: internal automation framework
Problem statement
Teams need AI workflows that can reason, retrieve, call tools, request approval, and recover from failure without disappearing into one large prompt.
Project implementation
Stack
Folder
Open project folderWhat it proves
You can model planning, tools, approvals, retries, and traces as explicit workflow state instead of hiding orchestration in one prompt.
Next upgrade path
Connect the orchestrator to real tool runtimes and migrate toward LangGraph-style execution.
Key commands
npm ci
npm run typecheck
npm run build
npm run smoke
npm run evalCore workflows
- intake a business task
- build or refine a plan
- call tools or retrieval nodes
- request approval for sensitive actions
- finalize a structured result
Milestones
- Define graph state and node contract
- Add planning and retrieval nodes
- Add tool node and approval node
- Add checkpoints, replay, and failure handling
Acceptance criteria
- workflow state is explicit and serializable
- approvals are first-class state transitions
- tool errors can be distinguished from model errors
- at least one run can be inspected from trace data
Starter implementation
Starter code is now available in projects/p03-agent-workflow-orchestrator. The current starter uses a local state machine, mock tools, approval gating, inspectable traces, and a minimal eval so learners can understand orchestration fundamentals before adding framework complexity.
Portfolio packaging
Share the graph diagram, run timeline, approval flow, and one failure recovery example.
Monetization path
This can evolve into a reusable workflow engine for operations, customer support, internal copilots, or agent platform consulting.