Solution: Building a Meeting Notes Analyzer

This lesson provides a complete walkthrough of the solution for the meeting notes analyzer exercise. We will break down the problem into its core components and implement each agent and orchestrator step by step.

The architectural approach

To solve this problem effectively, we will use a hybrid orchestration architecture. This approach is ideal because our problem has two distinct phases with different requirements:

1. Extraction phase: The first phase involves reading the raw meeting notes and extracting three independent pieces of information: the summary, the action items, and the key decisions. Because these tasks do not depend on each other, they are perfect candidates for parallel execution. We will use a ParallelAgent to run three specialist extraction agents simultaneously, which is the most efficient approach.

2. Finalization phase: The second phase involves compiling the extracted information into a single, polished document and saving it to a file. This task has a strict dependency: it cannot begin until all three extraction tasks are complete. Therefore, we will use a top-level SequentialAgent to ensure this step runs only after the parallel extraction phase is finished.

This combination of ParallelAgent nested within a SequentialAgent gives us the best of both worlds: the speed of concurrent processing for the independent tasks and the reliability of a guaranteed order for the dependent tasks.

Implementing the extraction agents

Our first step is to build the three specialist LlmAgent that will run in parallel. Each agent will be given a single, highly focused responsibility and will be configured to save its output to a unique key in the shared session state.

The summary agent