LlamaIndex

Did you miss Jerry Liu at the AI Engineer World's Fair last month? Catch his keynote here! Last year Jerry's talk was the most-watched video from the conference, so you won't want to miss him breaking down the future of knowledge assistants. https://lnkd.in/g6jFdzTb

Read how LlamaIndex helps SCALEPORT AI to ship 2x faster and land more clients! They say what LlamaIndex does for them: ⭐️ Accelerated development ⭐️ Better OCR with LlamaParse ⭐️ Flexible data handling https://lnkd.in/gcAGAiX7

How do you build a complex, multi-agent tree system to handle customer interactions? We've got a new open-source repo showing one way! Our multi-agent concierge system has a bunch of regular sub-agents, plus 3 "meta" agents: ➡️ A "concierge" agent which asks the user what they want to do, which can be a complex, multi-step process requiring multiple agents to perform tasks ➡️ An "orchestration" agent which figures out which sub-agent to handle the immediate next task ➡️ A "continuation" agent which looks at the broader context to decide if the user completed their original task or still needs more sub-agents to execute The overall system allows each sub-agent to focus on a specific skillset without having to come up with a giant, over-arching prompt for the entire system. We think it's a flexible, powerful pattern that you can use, so we've open-sourced it! Check out the full blog post: https://lnkd.in/dB34RCu6 Or head straight to the repo: https://lnkd.in/d88YN-Ne

A hidden release that makes LlamaParse way better at RAG over complex documents: we’ve made huge improvements to markdown-based table reconstruction - this allows you to parse very complex tables while making sure the rows/columns are well aligned! 🔨 Check out the results below (source image, before fix, after fix). Tutorial on how to build advanced RAG over your PDFs with lots of text and complex tables: https://lnkd.in/g5P2CMUA

Worked with AI but wondering why to use LlamaIndex? This new video is an introduction to LlamaIndex and in particular its agentic capabilities, covering: ➡️ What is LlamaIndex, including: ⭐️ Frameworks in Python and TypeScript ⭐️ LlamaParse service for parsing complex documents for LLMs ⭐️ LlamaCloud service for end-to-end enterprise RAG ⭐️ LlamaHub for free downloads of data connectors, LLM adapters, embedding models, vector stores, agent tools and more ➡️ Why you should use it, including: ⭐️Build faster ⭐️ Eliminate boilerplate ⭐️ Avoid early pitfalls ⭐️ Get into production ⭐️ Deliver business value ➡️ What can it do, including code examples of how to build: ⭐️ Retrieval augmented generation (RAG) ⭐️ World class parsing ⭐️ Agents and multi-agent systems https://lnkd.in/d_D3fenm

An underrated capability of sonnet-3.5 is that it’s really good at chart understanding 📊 - compared to gpt-4o it’s much better at inferring chart values into a structured table. Thanks to our brand-new LlamaParse release 💫 you can easily use SOTA multimodal models like sonnet-3.5 for document parsing and structuring, with added validation/scalability/reliability benefits behind our infrastructure. Check out our full tutorial below and example from the Llama 2 paper. Huge shoutout to Pierre-Loic Doulcet and Sacha Bron for the exciting new features. Additional releases: - Fast Mode: Run LlamaParse with our core text layout capabilities without OCR/models, for 0.1c a page.  - Improved Table Reconstruction: Fewer hallucinations in reconstructing complex tables. Results coming soon. Notebook: https://lnkd.in/dBzRNYYc LlamaParse: https://lnkd.in/g3UmUkcD

New LlamaIndex Webinar 🚨We're excited to cohost a workshop with the cofounders of Deasie (Reece Griffiths, Leonard Platzer, Mikko Peiponen) on improving RAG with advanced parsing and metadata. The data processing layer is one of the most important pieces to get right for RAG. This means that AI engineers need to make careful decisions in terms of parsing and transformations - including metadata extraction and chunking - in order to make sure that their e2e QA system is surfacing relevant results. This is a nice two-part workshop that demonstrates the following: the value of good parsing itself over complex documents, with LlamaParse the value of additional value of adding in metadata through Deasie's powerful automated labeling platform We show overall experimental results over research papers validating the combination of parsing + metadata for good performance. This Thursday 9am PT, check it out! https://lu.ma/ufx77ve8 Deasie: https://deasie.com/ LlamaParse: https://lnkd.in/g3UmUkcD

Build-your-own graph query algorithm with LlamaIndex + Mistral AI 🔎🕸️ You’re not limited to text-to-cypher or pure vector search when you build GraphRAG - you can combine both techniques or define your own custom graph query algorithm, as long as you have access to the underlying property graph store. Ravi Theja Desetty has a full video tutorial and notebook showing you how to do this - combine text-to-cypher + vector retrieval with Cohere reranking, and synthesize the final results with mistral-large. Video: https://lnkd.in/g43-W-FJ Mistral AI cookbook: https://lnkd.in/gW6yEqdj

The future of document RAG is multimodal RAG 🖼️📑. In our brand-new cookbook we highlight a new multi-modal RAG architecture for processing a slide deck (heavy in text, diagrams, charts, tables) using LlamaParse, LlamaIndex, and gpt-4o. At the core is a hybrid text/image chunk. In contrast to standard RAG pipelines that only index/synthesize over text chunks, our multi-modal RAG setup does the following:  1. For each page, both a) parse out the text, but also b) screenshot each page as a separate image. You can use standard OCR techniques but also multimodal models for extraction. 2. Create a hybrid chunk that contains the parsed text, with a file link to the saved image 3. Use text embeddings to embed the chunk (note: you can use image embeddings as well, but we find gaps here)  4. Retrieve relevant chunks by text embeddings. 5. During synthesis, load in the text and image. This combines the benefits of text parsing (well studied), with the visual recognition benefits of multimodal models. We have a full cookbook showing how you can leverage LlamaParse to do both the text/image extraction, and then set it up with gpt-4o for a full pipeline. Check it out: https://lnkd.in/gjaFvwYE

We're co-hosting an online hackathon this weekend, starting Friday 19th! Use Llama 3 from Meta to build your AI app, and win thousands of dollars in cash and credits and other fun prizes from us and our co-hosts Together AI Milvus and lablab.ai! Sign up here: https://lnkd.in/dGN_kTgN