Llamaindex Hackathon Resource Guide

LlamaIndex is an open-source framework for AI applications that utilizes the RAG approach. It includes features such as PDF parsing, hybrid search capabilities, custom prompts, and support for SQL/Pandas query engines. This versatile tool empowers developers to build advanced AI systems with enhanced search functionalities and data processing capabilities.

---

Beginners

• What is Retrieval-Augmented Generation (RAG)? What is any of this?

• We have libraries in Python and TypeScript; the Python lib has more features

• Installation and getting started:

• Python

• TypeScript

• I want to customize…

• LlamaHub is a collection of data connectors, agent tools and pre-written code

• We have notebooks that show how to use the tools on LlamaHub

• create-llama is a command-line tool that sets up a starter full-stack app for you

• Installation instructions

• We strongly recommend starting with this if you’re not sure where to start!

• The RAG CLI is a command-line tool that will index and query files for you

• Good for quickly checking if indexing documents works

• Building a low-code chatbot with Streamlit (this tool is called RAGs)

• What can I use?

• All the vector stores we support

• All the LLMs we support

• All the embedding models we support

• What’s the best way load and parse data?

• The ingestion pipeline

• How do I get full control over how my query engine works?

• The query pipeline

Intermediate

• We have several pre-built repos you may want to fork and modify:

• SEC Insights, an advanced full-stack RAG app using multi-document agents

• A full-stack demo repo using MongoDB Atlas

• A (very similar) demo repo using Azure Cosmos DB (a MongoDB clone)

• Create_llama_projects is a list of starter apps that can be installed with create-llama

• RAG over embedded tables

• Multi-document agent

• Multi-modal (images)

• Llamabot, a Slack bot

• Gives step-by-step instructions to build a Slack bot with memory

• LlamaParse, our advanced PDF parsing service

• There’s a special $500 + swag prize for best use of LlamaParse!

• Customizing prompts

• Hybrid search

• Vector search + keyword (BM25) search

Advanced

• These are mostly notebooks showing how we did it

• Customizing module types

• Every component — retrievers, parsers, query engines, agents — can be customized.

• Combining multiple data sources

• e.g. Slack + docs + API

• Routing across multiple sources

• Shows tool use, lets the LLM decide what tool to pick

• Multi-document queries

• Answering a single question that needs multiple sources to answer

• Examples of agents

• Autonomous decision-making using LlamaIndex

• Extracting structured outputs from unstructured data

• e.g. Turn web pages into JSON

• Multi-modal examples

• Interpreting images, retrieving images

• Auto-retrieval aka metadata filtering

• Get the LLM to pre-filter data by tag in your database

• Recursive retrieval

• Treat sub-nodes differently when retrieving

• Text to SQL

• Query a database using an LLM

• Pandas query engine

• Query ingested tables using Pandas