Part 2: Making Documents Searchable with Vector Stores and Retrieval

In this 3-part series, we walk through how to build a document intelligence app.

In Part 1, we explored Docling, a state-of-the-art tool for parsing and structuring documents—the first step in a RAG (Retrieval-Augmented Generation) pipeline.

In this part, we focus on a RAG workflow.

While there are many RAG tutorials available, this series shares an approach that has worked well for me, with a particular emphasis on content-rich documents such as images, tables, and text.

Let’s get started!

✅ Find the notebook for this article here: Language Models (Project 3)

Why should we combine an LLM with an external vector store?

Top-tier models can already analyze papers, financial reports, or news articles. They aren’t limited by their training cutoff anymore because they call web search agents that can pull up-to-date info from Google. You can even upload a few PDFs and chat with them directly. This works well for small tasks, but there are 2 main issues: