LlamaIndex Workflows

RAG agents need retrieval, reranking, synthesis, and validation as distinct steps. Workflows make each step testable and replace implicit chain logic with explicit event handlers.

LlamaIndex Workflows are a factory assembly line with sensors: each station emits a signal (event) when done, triggering the next machine.

Scenario

Query workflow: `QueryEvent` → retrieve chunks → `RetrieveEvent` → rerank → `SynthesizeEvent` → answer with citations.

Solution

In Agent Frameworks, apply LlamaIndex Workflows to this scenario: Query workflow: `QueryEvent` → retrieve chunks → `RetrieveEvent` → rerank → `SynthesizeEvent` → answer with citations. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to LlamaIndex Workflows (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where LlamaIndex Workflows happens in the code.

1from llama_index.core.workflow import Workflow, step, StartEvent, StopEvent, Event  # import dependencies2
3class QueryEvent(Event):  # define a data structure or component4    query: str5
6class AnswerEvent(Event):  # define a data structure or component7    answer: str8
9class RAGWorkflow(Workflow):  # define a data structure or component10    @step11    async def retrieve(self, ev: StartEvent) -> QueryEvent:12        return QueryEvent(query=ev.query)  # return the result13
14    @step15    async def synthesize(self, ev: QueryEvent) -> StopEvent:16        return StopEvent(result=AnswerEvent(answer="..."))  # return the result

• Treating LlamaIndex Workflows as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for llamaindex workflows

Quick recap — the most important points from this module.