Module 2 · Project 1: Business Operations Agent — Finance and Inventory (single agent with LangGraph)

OBJECTIVE: build an assistant for a small business that answers questions about sales, expenses and stock by querying its data, does calculations, suggests what to reorder, and creates purchase orders only after the owner approves — and learn how to build a reliable single agent with LangGraph. WHAT WE COVER (in project order): • Defining the product, its allowed actions and how we will judge it • Creating realistic business data (synthetic data) and a test set of tasks • Designing good tools: clear names, descriptions and safe inputs • Building the agent in LangGraph: state, nodes, edges, the tool loop • Memory: conversation threads and long-term facts • Human-in-the-loop approval before actions • Evaluation: task success, correct tool use, steps, cost; tracing with LangSmith or Langfuse • Deployment: Streamlit chat app AFTER THIS MODULE YOU CAN: • Build a tool-using agent with state, memory and approvals • Test an agent with a task set and read its traces to fix mistakes • Keep an agent safe and affordable with step limits and approvals DATA: synthetic sales, expenses, inventory and supplier tables you generate (SQLite). TOOLS: LangGraph, LangChain tools, SQLite, LangSmith or Langfuse, Streamlit.