Frequently Asked Questions
Frequently Asked Questions
Common questions about the workshop, what to expect, and how to prepare.
What makes a RAG pipeline suitable for multi-agent AI systems? +
A RAG pipeline designed for multi-agent AI needs several properties beyond basic retrieval: citation tracking that persists through agent handoffs, per-agent context management that respects individual agent context windows, shared memory that multiple agents can read and write consistently, conflict resolution for contradictory retrievals, and integration with MCP for typed retrieval requests and responses. This workshop builds all of these.
How do I add citations to a RAG pipeline for AI agents? +
The workshop covers citation tracking at every layer of the RAG pipeline — storing source metadata with embeddings, attaching source attribution to retrieved chunks, propagating citations through the LLM response generation, and verifying that claims in the final output trace back to specific retrieved sources. This citation chain makes the entire system auditable.
How does memory engineering work in a RAG pipeline for AI agents? +
Memory engineering in a multi-agent RAG pipeline involves three layers: working memory (the current agent context window), episodic memory (conversation history across sessions), and semantic memory (the embedded knowledge base). The workshop covers how to manage all three layers so agents have appropriate access to past interactions, structured knowledge, and current task context without context overflow.
What is the difference between RAG for chatbots and RAG for agent systems? +
Chatbot RAG is a stateless retrieval at each turn. Agent RAG must maintain state across agent handoffs, coordinate access to the knowledge base across multiple simultaneous agents, track which agent retrieved which information, and manage the context overhead that multiple retrieval operations add to each agent's context window. The workshop builds agent-grade RAG from the start.
How does this workshop prevent RAG hallucination in multi-agent systems? +
RAG hallucination prevention in this workshop operates at two levels: retrieval quality (using confidence scoring and source verification to surface only reliable documents) and generation quality (using citation-grounded prompting that requires the model to attribute every claim to a specific retrieved source). Uncited claims trigger a validation step before they propagate through the multi-agent system.
Can I use any vector database with the RAG pipeline from this workshop? +
Yes. The RAG architecture taught in this workshop is designed to be vector database agnostic. The retrieval interface is abstracted through MCP, so you can swap the underlying vector store without changing the agent orchestration layer. The instructor discusses vector database selection considerations for production RAG systems during the workshop.
