top of page

Agentic RAG: Qwen-3, CrewAI, and Firecrawl

  • Implemented an Agentic RAG application using CrewAI for agent orchestration, Firecrawl for web scraping, and a locally running Qwen-3 LLM, hosted on Lightning AI.
  • The system employs a two-agent workflow: a Retriever Agent that uses vector DB search and falls back to web search for context, and a Response Generator Agent that formulates the final response.
  • The Retriever Agent invokes relevant tools to gather context and generate insights, which are then used by the Response Generator Agent.
  • _Reactions_ highlight the importance of web search as a fallback mechanism in Agentic RAG and inquire about the performance comparison of Qwen-3 with other LLMs.
  • _Reactions_ also inquire about the planning aspects within the Agentic RAG, specifically whether reasoning is embedded in the prompt flow or separated into its own agent/module.
  • The workflow demonstration's visuals were well-received, with inquiries about the tool used for their creation.
Source:
bottom of page