🚀 Multi-AI Agents are now live! Craft your AI team: one agent researches, another executes tasks. They write, summarize, and edit—think of it as your mini-me doubling your productivity! 🤖 What would you build? Reply with your AI Agent idea for a chance to win free SWAG! ✨🐑

For folks looking to optimize their RAG using simple, proven approaches check out these talks, notes, slides we’ve made available for free from our course thanks to Jo Kristian Bergum Ben Clavié jason liu (More coming soon) parlance-labs.com/education/rag/

I thought a bit about what a “generally good” RAG pipeline looks like over complex documents (e.g. a PDF with tables/diagrams/weird layouts) 1. Parse document into a document graph of embedded objects (e.g. with LlamaParse) 2. Extract one or more text representations for each

Jerry Liu Agree. Document becomes structured lists of pages consisting of header, paragraph, table, figure objects. Each table and figure has an associated description drawn from the surrounding context. When building the retriever, parse paragraphs and descriptions into propositions

If you are looking for something to read this weekend, I am happy to share that Chapter 7 on instruction finetuning LLMs is now finally live on the Manning website: manning.com/books/build-a-… This is the longest chapter in the book and takes a from-scratch approach to implementing

Was positively surprised by Mistral AI's Mathstral release yesterday! I couldn't help it and wanted to give it a try. So, I just ported it to LitGPT, and it works really well based on my first impressions. Overall, it's case study for small to medium-sized specialized LLMs!

Join Manuel Faysse and me for the LlamaIndex Webinar this Friday, July 26th, at 9 AM PT! We'll be discussing our latest model for document retrieval: ColPali. Don't miss out! 👋🏼

This man can predict the future. He was an early investor in Uber, Twitter, and Notion. And he just said "In 50 years, everyone will be working for themselves." Naval Ravikant's 5 predictions on the future of wealth creation (and why you should care):

IR and NLP benchmarks are clean preprocessed text only. The problem is that real-world IR use cases don't have the luxury of preprocessed clean text data. Vision-LLMs offer a refreshing alternative that I strongly believe will shape the future of real-world IR. What You See

1/ The Bloom-Bilal Rule: When bored and lacking ideas, keep walking until the day becomes interesting. (via @sahilbloom+ Bilal Zaidi)

The secret is there is no secret.

The video of my Subaru Suresh Distinguished Lecture at IIT Madras yesterday is available. youtube.com/live/pjqKHOeyk…

Stanford CS229: Building Large Language Models This 1.5 hours lecture provides a concise overview of building a ChatGPT-like model, covering both pretraining (language modeling) and post-training (SFT/RLHF). For each component, it explores common practices in data collection,

Kicked off the 2nd batch of Turing’s Dream, the AI residency that I run in Bangalore! Here’s what they’re upto…

I'm starting something soon. Join me in making state-of-the-art AI models for the world from 🇮🇳India.

may have hit a nerve here...it all started with trying to understand how production AI systems actually work FOLKS - I've tried every agent framework out there and talked to many strong founders building impressive things with AI BUT I was surprised to find that most successful

LMAO 😂 Vibe-coding is fun until you have to debug it without a clue.

Sonnet 4.0 is cracked at document understanding. With our latest update, we’ve built a sonnet 4.0-powered agent to help convert the most complex docs into markdown, detect layouts and tables/images. The agent loop helps prevent hallucinations and join tables across multiple

Why your RAG system is failing despite "great" embedding scores I just watched Kelly Hong from Chroma present their research on generative benchmarking, and it's a wake-up call for anyone building retrieval systems. The uncomfortable truth: your embedding model might be

When building RAG systems, think like a data scientist not an engineer. Extract patterns, identify value, then build the simplest thing that works. I've seen teams waste months on complex architectures when the real solution was just better metadata filters.