← Back to all workshops

Workshop 7: LangFlow

Building an augmented search engine with RAG

What is LangFlow?

LangFlow is a visual tool for building LLM workflows. Instead of writing code, you connect components together like building blocks. In this workshop you'll build a RAG (Retrieval-Augmented Generation) system using Astra DB.

What is RAG?

RAG lets an LLM answer questions about documents it wasn't trained on. It works in two steps:

1. Retrieval — find relevant chunks from your documents
2. Generation — use those chunks as context to answer the question

This is how you make an LLM "read" a PDF, a website, or any custom data.

The Tool

I will provide you with an url for the LangFlow instance.

Exercise: IPCC Report Q&A

Build a search engine that can answer questions about the latest IPCC climate report.

Step by step

1. Open LangFlow and create a new project
2. Select the RAG template from the templates
3. Upload the IPCC report PDF
4. Configure the components:
   • Document loader — to read the PDF
   • Text splitter — to chunk the document
   • Embeddings — to convert chunks to vectors
   • Vector store — to store and search chunks
   • LLM — to generate answers
5. Run the flow and test with questions

Test questions

Try asking your RAG system:

• What is the projected global temperature rise by 2100?
• What are the main causes of sea level rise?
• Which regions are most vulnerable to climate change?
• What mitigation strategies does the report recommend?
• What is the carbon budget to stay below 1.5°C?

Verify: Check the retrieved chunks. Does the answer actually come from the document, or is the LLM making things up?

Experiment

• Change the chunk size — what happens with smaller or larger chunks?
• Try a question that isn't in the document — does the system admit it doesn't know?

Questions to Consider

• Why do we need to split the document into chunks?
• What happens if a relevant answer spans two chunks?
• How is this different from just pasting the whole document in the prompt?
• When would RAG fail? What are its limitations?
• How could you evaluate if your RAG system is working well?