/ module 06

Orchestration

A bare LLM is a text-in, text-out function. To build real applications you wrap it in a flow: routers decide what to do, tools take action, retrievers fetch knowledge, and memory persists across turns. This is what frameworks like LangChain and LlamaIndex are for.

patterns

Components of an agent

Router classifies intent → pick a tool. Tools are typed functions the model can call. Memory stores past turns or facts. A loop lets the LLM call tools repeatedly until done.

tool / function call
Structured output the LLM emits to trigger an external action.
ReAct loop
Reason → Act → Observe → repeat until the agent answers.
guardrails
Validation around inputs and outputs (PII, schema, safety).
Live lab · Compose & run an agent
Preset:
Flow (click a node to remove, or add below)
input
answer
+ add:
Trace
(run the flow)
Final answer
(waiting)