FRS102 RAG
This flow uses two agents and is a RAG (retrieval augmented generation) workflow
The first agent performs a web search for FRS102 documentation and proposed updates.
It then uses a document loader and text splitter to chunk the documents before loading them into a Pinecone vector database
The second agent starts with a webhook trigger. Once this trigger is activated a call is made to an LLM that looks for relevant chunks in the vector database and sends the question and document chunks to the LLM for a response. It then ends with a webhook.
The trigger and response webhook is used by an ElevenLabs voice agent. The user can speak their questions to this ElevenLabs voice agent and it will respond audibly as well.
This second agent uses simple memory so that it retains knowledge of the conversation so that if the user asks a follow up question the agent knows the context of the conversation