Introduction to Query Pipelines (Building Advanced RAG, Part 1)

This episode gives you an introduction to using Query Pipelines in LlamaIndex to setup basic to advanced orchestration over your data. This ranges from simple prompt chains to full DAGs that contain a lot of advanced retrieval components.

This video mini-series around advanced RAG orchestration is designed to help you get a handle of how to build complex workflows over your data for different use cases.

EDIT: at 26:19 we erroneously mention `FunctionComponent`, which doesn't exist. Try `FnComponen` instead!

Timeline:
00:00-07:00 - Intro
07:00-15:50 - Prompt Chaining
15:50-24:13 - RAG Workflows
24:13-27:50 - Custom Components
27:50 - Async/Parallel

[AI__Spectrum]

OMG THIS SERIE IS JUST It's truly incredible! We are waiting for more Multi-Agent videos :)

[AngusLou]

Smart ways and innovative ideas to re-invent RAG, good job

[nigerianprince]

lost my fire for LLMs and RAG in the last few months as even though i can be technical, it took to much time for me to learn, understand and implement while working full time.

this video rekindled that fire because it's much easier than i remember it to build a RAG Pipeline. Keep up the great work !

[geoffreygordonashbrook]

Many thanks for the tools and videos! Request: could you sometime cover in extra detail how to locally run locally downloaded (and locally made / fine tuned) models as part of RAG systems? Being able to swap in and out local models (and fine tune those) is crucial in many use-cases. But available guides nearly all use a non-modular workflow where either a cloud-api mode or a fresh-hugging-face download are required, neither of which are suitable for many production or even research cases. Many thanks!

[bwc]

Thank you for the video... here I am the next morning ~2M tokens in to a test use case (thanks for the phoenix tracing tip too..)

[unclecode]

Many thank fro this video series, how to get the output of each module individually after running the chain for inspection. Thx

[nikhilshrestha]

Great video. Just a few questions:
1. Does the speed of this system remain so much ?
2. I tried the same by including 40MB and 20 MB book with small kbs and mbs PDFs. The speed was somewhat similar.

[HaiNanZhifa]

Hi Jerry, when I have a corpus of 100 million legal documents, what are the effective indexing strategies available? How can I apply the techniques you showed here and in your other videos?

[SuperLiberty]

Hey! I'm looking for effective strategy how to handle structured data. Could you advise tutorial or paper to study? Regards!

So many thanks for your hard work guys!! I'm following most of the guides but there is something that's killing me. Just imagine loading a document with "chapters", "subchapters" and, obvioulsy a parragraph inside... RAG will work flawlessly extracting chunks of these parragraphs but if you ask for a resume of all the subchapters of certaing chapter then the RAG don't work so well if the context is huge (due missed chunks).
¿There is any kind of metadata/chunking way of for this kind of questions?
For more context, I'm using Sherpa for reading a PDF. The PDF has a list of chapters about certain things. The "query" is, "what's included in this chapter" and the expected response is a list of all subchapters... Obtained response is not complete due lack of context.

[thetagang]

I know the data used for the code is PG essays, but there's quote a few of them.... is there somewhere to download this??

[zugbob]

Is this all possible in the TS version?

[navanshukhare]

Hi, Can you please make an advanced video on different types of engines, especially SQLAutoVectorQueryEngine? A lot of features and we love the llama index frameowrk

[ajg]

I'm curious about something here. As we're generating indexes from a file, it's possible to create diverse types of indexes, such as VectorStoreIndex, Summary Index, Tree Index, Keyword Table Index, and Knowledge Graph Index, etc ...

What are the advantages of utilizing these various indexes? How can we determine which indexing method is most suitable for our specific use case?

[ndamulelosbg]

So basically this is a more generalsed 'chain' concept commonly used in langchain?

[austinmw]

What happened to FunctionComponent? I don't see it in the latest LlamaIndex version 🧐

[sritharan]

Hey Jerry, first off llama_index is awesome, pls revisit or change the docs side tree panel, its very annoying, same doc is referenced in multiple places and it keeps jumping around, a good example will be langchain docs, but i love using llama_index. thx

[iloos]

differrence between this and langgraph?