Introduction

This project implements a combination of SequentialAgent and LlmAgent ADK agents, as following:

• Retriever Agent - LlmAgent - calls Vertex AI RAG to fetch contextually relevant passages from a structured corpus.
• Analyzer Agent - LlmAgent - interprets and summarizes retrieved text, highlighting key points, conflicts, and evidence.
• Final Answer Agent - LlmAgent - constructs a human-readable response with inline citations ([P1], [P2]), grounding every claim in the retrieved evidence.
• root_agent - SequentialAgent - orchestrate the 3 LlmAgent agents.

Create the corpus and ingest data

Follow the steps:

1. Copy your data files in a GCP storage bucket, identified in the .env file as GCS_URI.
2. Set your GOOGLE_CLOUD_PROJECT, GOOGLE_CLOUD_LOCATION parameters in your .env file.
3. Set also GOOGLE_GENAI_USE_VERTEXAI to True.
4. Run:

   python adk-rag-agent/create_database/create_corpus_and_vector_database.py

   This will initialize the Vertex AI client, will create the corpus, download and upload to corpus the data.

Running the Agent

You can run the agent using the ADK command in your terminal. from the root project directory:

1. Run agent in CLI:

   adk run adk-rag-agent

2. Run agent with ADK Web UI:

   adk web

   Select the adk-rag-agent from the dropdown and start interogating the agents.