Agentic RAG using Vertex AI Playbook or Google Products?

[Deepak_Mallya]

Hi everyone,

I’m exploring how to send a user query to multiple agents, each with their own data stores, consolidate the information using a Language Model (RAG), and then provide a response to the user query. Has anyone implemented a similar setup using Vertex AI Playbook or other Google products? Any insights or experiences would be greatly appreciated, especially regarding the setup process, challenges, and best practices.

Thanks in advance for your help!

[pedropcamellon]

Hi. What have you tried so far?

[Deepak_Mallya]

Currently, I use a "Steering" playbook that acts as a gateway for all requests. I forward the request to "Playbook A," which responds to the user's query. I then pass it to "Playbook B," which does the same. What I need is to consolidate all the answers provided by these agent playbooks, summarize them in the steering playbook, and then provide that response to the user.

[pedropcamellon]

Have you tried telling "Playbook B" to return the input and its output?

[Deepak_Mallya]

Yes. I have been trying this with little success. Is there any documentation link that can help me?

[pedropcamellon]

I think it can be solved using a good prompt and examples. Something like ``` You will get a question from the user. Return the question and your answer...``` then set two or three examples of how the flow should look. Check this channel (URL Removed by Staff). It has good videos. Unfortunately, I have not been able to find too many docs about this platform. If you find some please share the links.