LlamaIndex’s Post

One of the best indicators of a data science team's knowledge and maturity around LLM usage is the level of attention they are currently giving to upfront data preparation. 📣 Tomorrow, we're excited to be teaming up with Jerry Liu and the team from LlamaIndex to host a webinar on best practice data preparation for RAG workflows. We'll be showing how their LlamaParse solution for intelligent parsing and document transformation can be combined with Deasie's metadata creation workflow to build accurate and scalable knowledge retrieval over complex documents. When: Thursday 18th July, 9am PT / 12 ET Where: https://lu.ma/ufx77ve8 Who: Relevant for both AI engineers as well as company leaders looking to build their data foundation for GenAI

Join us June 25th at the AIQCon conference. Jerry Liu, CEO & Co-Founder of LlamaIndex will be giving his talk "Building Advanced Question-Answering Agents Over Complex Data" Large Language Models (LLMs) are revolutionizing how users can search for, interact with, and generate new content, leading to a huge wave of developer-led, context-augmented LLM applications. Some recent stacks and toolkits around Retrieval-Augmented Generation (RAG) have emerged, enabling developers to build applications such as chatbots using LLMs on their private data. However, while setting up basic RAG-powered QA is straightforward, solving complex question-answering over large quantities of complex data requires new data, retrieval, and LLM architectures. This talk provides an overview of these agentic systems, the opportunities they unlock, how to build them, as well as remaining challenges. Make sure to get your tickets now before it's too late! https://lnkd.in/dR87sR7A

Just finished the course “Vector Databases in Practice: Deep Dive” by Joon-Pil Hwang! Check it out: https://lnkd.in/dGEapvc6 #vectordatabasedevelopment #vectordatabases #RAG #Gen-AI

Interested in incorporating on-device AI into your projects? Check out this informative video on Getting Started with On-Device AI using ObjectBox Vector Database and LangChain:

pykoi is an open-source data curation platform for LLMs. Using pykoi, everyone can build robust language models through faster data curation using both human and machine feedback. It provides support for each step in the MLOps cycle, from data labeling, model monitoring, to RLHF. Using pykoi, one may easily: Collect data and feedback from the users of your language models Compare the performance of models to each other Apply reinforcement learning with human feedback to your models Create data labeling and annotation tasks for your own users Finetune your model to personal files Understand how users are using your language models Github - https://lnkd.in/g5HnNsyj Docs - https://lnkd.in/gwhrPbd6

Hugging Face has an open-source Cookbook with many applied AI recipes 📖 Here are some of the latest recipes contributed ⥥ - "Information Extraction with Haystack (deepset) and NuExtract": Use Haystack and transformers to build structured data extraction pipelines using LLMs by Stefano Fiorucci - "Build RAG with Hugging Face and Milvus": Learn how to use Milvus with sentence transformers to build RAG pipelines - "Code Search with Vector Embeddings and Qdrant": Learn how to search a codebase through retrieval using Qdrant, sentence transformers and Jina AI code embeddings - "Agentic RAG: turbocharge your RAG with query reformulation and self-query": Boost your RAG pipeline by swapping it with agents by Aymeric Roucher Link to all of them are in comments 💬

RAG - Retrieval Augmented Generation - is now the established approach to enterprises bringing LLM to their down data. We are seeing a rapid progression from naive RAG to advanced RAG as enterprises try to tame the predictability, quality, speed, and cost of LLM responses. Improvements are happening at every step: - Routing queries in a multi-model RAG workflow - Enrichment with Metadata - Selection of relevant vectors during retreival - Evaluating answer quality with another LLM We have documented some of the best practices for RAG in this in depth article. https://lnkd.in/g6G4SpTZ

Congratulations to Qiming (Bill) Bao and our other co-authors (listed in Qiming's post) on the acceptance of this paper on improving abstract reasoning via training augmentation (evaluated on the ReClor dataset). This is part of our multi-year effort to improve Deep Learning-based reasoning, at the Strong AI Lab (SAIL) in the NAOInstitute at School of Computer Science - University of Auckland. We'd love your support and collaboration in this effort, and always encourage extremely strong students (like Qiming) with strongly related research goals to apply to join us for their PhD. Solving Reasoning within LLMs and similar systems was recently mentioned by Sam Altman at Open AI as one of the biggest problems that needed to be solved. Perhaps you can help us work on it. Kai-Fu Lee , Dario Gil , Eric Xing , Mustafa Suleyman , Christopher Manning , Bowen Zhou , Yi Zeng , Aya Soffer Lior S.

Did you know that we have a newsletter devoted to AI news, funding, research, and other recent updates? Sign up for our AI Weekly Recap newsletter here! #DataScience #AI #ArtificialIntelligence https://hubs.li/Q02GMP-p0

🦄 Single Agent multi-documents vs Multi Agent multi-documents RAG framework. Till now in Langchain, I have experimented with having a single Agent invocation chain to gather thoughts, observation and action + running those thoughts on AI search (one index) + response generation. This one, though it directs the (may or may not)distracted question from user towards its intent, its ultimately on the search system(improved precision) to not miss any information matching the Agent formed question. Having multiple agents for each document that needs RAG framework in. Maintaining two types of indices - summary + vector index for each document. Having an Index node that encompass the definition of what each Agent can do. Retriever is acted upon on these Index nodes for each document/Agent. These REACT prompting based framework is cool on the first part of RAG - query formation/comprehension - there is still a gap to be bridged by adding a separate layer (topics retrieval, scoring profiles, semantic closeness ranking) to enhance the retrieval + response aspect of RAG.