[Hands-on] Enterprise-grade Agentic RAG with EyeLevel and CrewAI

Vanilla RAG works as long as your external docs. look like the image on the left, but real-world documents are like the image on the right:

Today, let’s build an enterprise-grade agentic RAG over such complex real-world docs.

Here’s our tech stack:

• CrewAI for agent orchestration.

• EyelevelAI’s GroundX for SOTA document parsing.

We’ll do RAG over MIG 29 (a fighter aircraft) flight manual, which includes complex figures, diagrams, and more.

This diagram illustrates the key components & how they interact with each other!

Let’s build it!

---

Setup the LLM

CrewAI nicely integrates with all the popular LLMs and providers out there!

Here's how we setup a local DeepSeek-R1:7B:

Setup tools

This is the core of our application. We integrate EyelevelAI’s Python SDK as a custom tool with CrewAI.

Define Retriever Agent

The retriever agent is responsible for retrieving the right context for the user query and is assigned a task. Here's how it's done:

Define Response Gen Agent

The Response Gen Agent takes the query and context provided by the retriever agent and generates a response.

Setup Crew

Once we have the agents and their tasks defined, we put them in a crew that is orchestrated using CrewAI.

Kickoff and results

Finally, we provide the user query and kick off the crew!

And it works as it should, even on a complex document, which is impressive, isn’t it?

You can test EyeLevel’s SOTA document parsing on your own complex document here →

They are developing powerful parsing systems that can intuitively chunk relevant content and understand what’s inside each chunk, whether it's text, images, or diagrams, as shown below:

As depicted above, the system takes an unstructured (text, tables, images, flow charts) input and parses it into a JSON format that LLMs can easily process to build RAGs over.

Try EyeLevel to build real-world robust RAG systems:

Find the code for this demo here: Agentic RAG with EyeLevel and CrewAI.

Thanks to EyeLevel for showing us their powerful RAG solution and partnering on today’s issue.

Thanks for reading!