Skip to content

Question-Answering (RAG)#

One of the most common use-cases for LLMs is to answer questions over a set of data. This data is oftentimes in the form of unstructured documents (e.g. PDFs, HTML), but can also be semi-structured or structured.

The predominant framework for enabling QA with LLMs is Retrieval Augmented Generation (RAG). LlamaIndex offers simple-to-advanced RAG techniques to tackle simple-to-advanced questions over different volumes and types of data.

There are different subtypes of question-answering.

RAG over Unstructured Documents#

LlamaIndex can pull in unstructured text, PDFs, Notion and Slack documents and more and index the data within them.

The simplest queries involve either semantic search or summarization.

  • Semantic search: A query about specific information in a document that matches the query terms and/or semantic intent. This is typically executed with simple vector retrieval (top-k). Example of semantic search
  • Summarization: condensing a large amount of data into a short summary relevant to your current question. Example of summarization

QA over Structured Data#

If your data already exists in a SQL database, CSV file, or other structured format, LlamaIndex can query the data in these sources. This includes text-to-SQL (natural language to SQL operations) and also text-to-Pandas (natural language to Pandas operations).

Advanced QA Topics#

As you scale to more complex questions / more data, there are many techniques in LlamaIndex to help you with better query understanding, retrieval, and integration of data sources.


LlamaIndex has a lot of resources around QA / RAG. Here are some core resource guides to refer to.

I'm a RAG beginner and want to learn the basics: Take a look at our "Learn" series of guides.

I've built RAG, and now I want to optimize it: Take a look at our "Advanced Topics" Guides.

I want to learn all about a particular module: Here are the core module guides to help build simple-to-advanced QA/RAG systems:

Further examples#

For further examples of Q&A use cases, see our Q&A section in Putting it All Together.