Jerry Liu
Sunnybank Hills
Is this you? As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.
Claim your profile
Articles by Jerry Liu
Just-in-Time Agentic OCR
The latest RAG trend I'm seeing in the current agent harnesses (Claude Cowork, Codex) is to do two passes of document processing in order to solve a knowledge work task over a data room of documents: A fast and light pass, oftentimes using a free/OSS doc parsing tool. This can be cheaply run across 10-100 files, and lets the agent do retrieval (grep, semantic search) to find the relevant subsets of context. A "just-in-time" VLM-based pass.
An Electronic Health Record–Integrated, Large Language Model–Powered Tool to Triage Surgical Patients
An Electronic Health Record–Integrated, Large Language Model–Powered Tool to Triage Surgical Patients Supplement 1. eAppendix 1. End-to-End Clinical Workflow for SCM Triage and Perioperative Comanagement eAppendix 2. Architecture and User-Facing Workflow for SCM Navigator eAppendix 3. SCM Navigator Prompts eAppendix 4. Physician Training Handout eFigure. Forest Plot of Adjusted Odds Ratios for Correct SCM Navigator Classification by Demographic Subgroup eTable 1.
Document OCR is Not Getting Commoditized
The standard take on document parsing is that it's a melting ice cube. The reasoning is the following: every frontier model release reads PDFs a little better, so within a couple of model generations OCR becomes a checkbox feature of the model API, and paying a specialized provider for it will look silly. We think the opposite is happening, and the benchmark data agrees.
Modeling the Seasonal Dynamics of Surface Water Networks Using DEM
You are already at the latest version The topology and spatial extent of regional surface-water bodies fluctuate seasonally, with the most pronounced changes during high-intensity rainfall and flooding. To capture these dynamics, we introduce a novel geomorphological metric, the catchment-to-destination area ratio (C/D ratio).
Open source tools that help you get promoted
Hey Product Hunt ğ I'm Jerry, founder of Dayflow. Here's a bitter lesson that everyone learns first-hand: the person who gets promoted isn't the one who did the best work. It's the one who can remember it and come with receipts. You know the feeling. Your manager asks "what did you ship this quarter," and your mind goes blank.
Build Automated Loan Income Verification with LlamaParse + Claude Agent SDK
Mortgage loan processors spend a surprising amount of time on what should be mechanical work: pulling numbers out of documents and checking that they match. According to industry estimates, manual document handling consumes 40-60% of a loan processor's time. The average mortgage file is 500+ pages across dozens of document types — tax returns, pay stubs, bank statements, W-2s, the 1003 application, appraisals, title docs, and more. The core challenge is income consistency.
Improving Table Parsing for Word (.docx) Documents
Word docs are one of the most common file formats we process in LlamaParse, and they've always been surprisingly frustrating to parse well. The reason is genuinely counterintuitive: the .docx format has better structural information than most document formats, but until now we haven't been able to fully use it. We just shipped an update that changes this for tables.
OCR for Invoices: How to Extract Data with Accuracy and Speed
Digital transformation has reached nearly every department in modern organizations. Marketing runs on automation. Sales operates in CRMs. Operations rely on real-time dashboards. And yet, in many finance departments, invoice processing still runs on manual data entry.
What Is Agentic OCR? The Next Evolution of Intelligent Document Automation
An invoice arrives. Your OCR system scans it, pulls out the line items, and dumps a blob of text into your ERP. Then the total doesn't match the purchase order. Turns out the vendor changed their invoice layout this month, with two columns where there used to be one. Your OCR system missed the subtotals. Someone finally catches the discrepancy three days later, after the payment has already been processed. This scenario plays out thousands of times a day across enterprises.
OCR for Accounts Payable: Benefits, Challenges, and Best Practices
Accounts Payable teams are under constant pressure to process invoices faster while maintaining financial accuracy and compliance. As organizations scale, invoice volumes increase, vendor formats diversify, and approval workflows grow more complex. What often appears to be a simple data entry function quickly becomes an operational bottleneck. Manual invoice processing carries hidden costs. Every minute spent entering data from paper invoices or PDF attachments is time diverted from meaningful work.
LlamaIndex is more than a RAG Framework. It is Agentic Document Processing.
We launched LlamaIndex as an open-source framework in late 2022, right at the start of the RAG wave. The framework grew fast (47K GitHub stars, 5.2M monthly downloads), and became a core starter toolkit for any developer building RAG and agentic applications in production. Since then, the AI orchestration landscape has drastically evolved.
OmniDocBench is Saturated, What’s Next for OCR Benchmarks?
We’ve seen a whirlwind of advancements in document OCR VLMs in the past month, from Deepseek-OCR2 to PaddleOCR-VL-1.5 to GLM-OCR. The latest model, GLM-OCR, set a new SOTA benchmark on OmniDocBench v1.5 with 94.6% accuracy, beating not only OSS models but also recent frontier models like Gemini 3 Pro and GPT-5.2. OmniDocBench has emerged as one of the default benchmarks for evaluating any model on document understanding.
Long Horizon Document Agents
Recently Sequoia predicted that 2026 is the year of long-horizon agents. Long horizon agents are “AI that can figure things out [and] has some baseline knowledge (pre-training), the ability to reason over that knowledge (inference-time compute), and the ability iterate its way to the answer.” At LlamaIndex, we’re especially excited about how this translates into more end-to-end knowledge work automation, particularly over documents.
Files Are All You Need
We’re seeing an interesting trend where the primary way agents are able to read and interact with unstructured context is through files. This blog paints an overview of the core interface, where they’re being used, and questions we still need to explore going forward.
Six Big Bets
You don’t get to take unlimited big bets in a single lifetime. You get a small number of windows where risk tolerance, energy, capital, and conviction align. In The Black Swan Nassim Taleb frames this as asymmetric exposure to unlimited upside with limited downside. And startups are one of the clearest modern mechanisms for that kind of asymmetry. But what’s often missed is that these opportunities are finite. You can’t swing endlessly.
OlmOCR-Bench Review - Insights and Pitfalls on an OCR Benchmark
Document OCR has been around for decades, but it has traditionally relied on complex hand-written pipelines and “traditional” ML systems. Over the past two years we’ve seen huge advancements in document understanding powered by general frontier models and specialized models. Models like dots.OCR, PaddleOCR, DeepSeek-OCR, and OlmOCR/OlmOCR2 are helping to drive this “brave-new world” of general visual understanding across an increasing long-tail of document types.
[2512.00207] Constructing Efficient Fact-Storing MLPs for Transformers
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Adding Document Understanding to Claude Code — LlamaIndex - Build Knowledge Assistants over your Enterprise Data
The rise of coding agents like Claude Code, Cursor, Windsurf, Cognition, Lovable, etc. marks a shift in how software is built. Instead of manually wiring APIs together, you can describe what you want through natural language, and the agent can handle the technical task of writing, executing, and iterating on the code. This opens up the possibility for “low-code IT” and allowing business users to quickly build internal and external-facing applications.
LlamaCloud One Year Later: The Complete Document Automation Platform — LlamaIndex - Build Knowledge Assistants over your Enterprise Data
Hi everyone, Jerry here, CEO and co-founder of LlamaIndex. It's been an incredible year since we first launched LlamaCloud in public beta. As I reflect on this journey, I keep coming back to a fundamental truth I've witnessed since the early days of LlamaIndex as a developer framework, from hundreds of Discord messages to F500 enterprise conversations: the world runs on documents. Walk into any company, and you'll find the same story.
Advance care planning and hospice utilization among Japanese Americans: a scoping review
Although advance care planning (ACP) has increased in recent years, there is still a discrepancy in the utilization of these resources among the Asian American, Native Hawaiian, and Pacific Islander populations. We present the first scoping review of ACP and hospice utilization among Japanese Americans. This scoping review was created in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020. A literature search was conducted using PubMed.
Colorectal Disease | ACPGBI Gastroenterology Journal | Wiley Online Library
This study affirms the value of rectal cancer multidisciplinary conferences (MDCs) by demonstrating that they change 40% of patient care plans. Radiology reports change 25% of the time and are frequent drivers of plan alterations. The study bolsters MDC review of all rectal cancers as no single factor predicts plan confirmation. INTRODUCTION The optimal treatment for rectal cancer often requires multimodality therapy.
Build powerful RAG pipelines with LlamaIndex and Amazon Bedrock
Starlink 2 hours ago After seeing Wi-Fi network named “STINKY,” Navy found hidden Starlink dish on US warship Ars Technica - Nate Anderson • 2h To be fair, it's hard to live without Wi-Fi. It's no secret that government IT can be a huge bummer.
Small Business Meetup
More About Small Business Meetup What to expect: Relaxed and friendly atmosphere: We foster a welcoming environment where everyone feels comfortable to participate and network. Short, impactful presentations: Each month, we feature valuable presentations led by experts on relevant marketing and business topics. Ample networking time: Connect with fellow entrepreneurs, share experiences, and build valuable relationships.
Small Business Meetup
Each month, we feature valuable presentations led by experts on relevant marketing and business topics. Special Guest Speakers:Kim Amor and Nick Hosking Your Local SEO Experts, KN Website Design In this session, you'll learn how to harness the power of SEO to not only boost your online presence but also forge valuable connections and fuel your company's growth.
Effects of Diet and Antibiotics on Anastomotic Healing: A Mouse Model Study with Varied Dietary Fiber and Fat, and Pre-operative Antibiotics.
This study investigated the separate impacts of diet and pre-operative antibiotics on gut microbiome and colonic anastomotic healing using a mouse model. Male C57BL/6J mice were fed either low-fat-high-fibre (SD) or high-fat-low-fiber (WD) groups for 6 weeks, then further received either pre-operative antibiotics or a control sham before a colonic anastomotic procedure was performed.
Small Business Meetup
Each month, we feature valuable presentations led by experts on relevant marketing and business topics. Special Guest Speaker: Dean Munro, Business Advisor. Self-employment Assistance Program. ETC Ltd. Join us for an informative session led by Dean Munro, Business Advisor at the Self Employment Assistance Program of ETC Ltd.
Introducing LlamaCloud and LlamaParse
Today is a big day for the LlamaIndex ecosystem: we are announcing LlamaCloud, a new generation of managed parsing, ingestion, and retrieval services, designed to bring production-grade context-augmentation to your LLM and RAG applications. Using LlamaCloud as an enterprise AI engineer, you can focus on writing the business logic and not on data wrangling. Process large volumes of production data, immediately leading to better response quality.
Introducing LlamaCloud and LlamaParse
Today is a big day for the LlamaIndex ecosystem: we are announcing LlamaCloud, a new generation of managed parsing, ingestion, and retrieval services, designed to bring production-grade context-augmentation to your LLM and RAG applications. Using LlamaCloud as an enterprise AI engineer, you can focus on writing the business logic and not on data wrangling. Process large volumes of production data, immediately leading to better response quality.
An Attribution Theory Perspective
The negative consequences of racial discrimination for both individuals and society are well documented. In this study, we focused on how workplace incivility (e.g., being ignored, ostracized or ridiculed by a colleague) relates to perceived interpersonal racial discrimination and addressed the following research questions: Do employees of color (Black, Hispanic or Asian) experience more (or fewer) uncivil behaviors at work, compared with white employees?
RAG is a hack - with Jerry Liu from LlamaIndex
Oct 5 • 1HR 8M Audio playback is not supported on your browser. Please upgrade. Appears in this episode Writes Jerry Liu Subscribe The podcast by and for AI Engineers! We are the first place over 50k developers hear news and interviews about Software 3.0 - Foundation Models changing every domain in Code Generation, Computer Vision, AI Agents, and more, directly from the founders, builders, and thinkers involved in pushing the cutting edge.
A Vector Store Walkthrough
from llama_index.schema import TextNode from llama_index.vector_stores.types import ExactMatchFilter, MetadataFilters nodes = [ TextNode( text="Not aware that tables of these squares existed, Bouvard, who calculated the tides for Laplace, underwent the labour of calculating the square of each individual sine in every case in which it occurred.", metadata={ "topic": "math", "entities": "laplace", }, ), TextNode( text="Within the limits of the lunar orbit there are not less than one thousand...
Data Agents
Today we're incredibly excited to announce the launch of a big new capability within LlamaIndex: Data Agents. Perform automated search and retrieval over different types of data - unstructured, semi-structured, and structured. Calling any external service API in a structured fashion. They can either process the response immediately, or index/cache this data for future use. Storing conversation history. Using all of the above to fulfill both simple and complex data tasks.
Better Enabling Bottoms-Up LLM Application Development
We created modular storage abstractions (data, indices), and compute abstractions (retrievers, query engines). We created a lower-level API where users could use our modules (retrievers, query engines) independently and customize it as part of a larger system.
Llama Index & Prem AI Join Forces
Co-authors: Simone Giacomelli (co-founder at Prem), Jerry Liu (co-founder/CEO at LlamaIndex) We're pleased to share the successful integration of Prem App and Llama Index, a union that brings a new level of privacy to AI development. Prem's self-hosting AI models and Llama's versatile data framework enhances the ability to build AI applications in a customizable and flexible manner.
LlamaIndex and Weaviate
Co-authors: Jerry Liu (co-founder/CEO of LlamaIndex) Erika Cardenas (Developer Advocate, Weaviate) While large language models (LLMs) like GPT-4 have impressive capabilities in generation and reasoning, they have limitations in terms of their ability to access and retrieve specific facts, figures, or contextually relevant information.
Building the data framework for LLMs
Today is an exciting day for LlamaIndex, and a big milestone in my personal journey with generative AI. I've followed generative models for most of my academic/professional career - from my research on GANs/sensor compression to following Transformers/ GPT developments. It became increasingly clear that as these models got bigger/better, they were evolving from knowledge generators to intelligent engines that could reason/act over new information.
Combining Text-to-SQL with Semantic Search for Retrieval Augmented Generation
In this article, we showcase a powerful new query engine ( SQLAutoVectorQueryEngine ) in LlamaIndex that can leverage both a SQL database as well as a vector store to fulfill complex natural language queries over a combination of structured and unstructured data. This query engine can leverage the expressivity of SQL over structured data, and join it with unstructured context from a vector database.
Dumber LLM Agents Need More Constraints and Better Tools
In this article, we compare how well LLM-powered agents with different degrees of complexity perform over practical data tasks (financial analysis). We compare the performance of agents with more complex, unrestrained interaction behavior (ReAct) with agents that contain simpler, more constrained interactions (routing). We specifically analyze how much complexity can be added to the agent layer vs. the tool layer. We find that the choice of the language model matters a lot.
Build a ChatGPT with your Private Data using LlamaIndex and MongoDB
Co-authors: Prakul Agarwal — Senior Product Manager, Machine Learning at MongoDB Jerry Liu — co-founder at LlamaIndex Update (6/22/2023): The preferred way to use LlamaIndex + MongoDB is now with our MongoDBAtlasVectorSearch class. Take a look at our guide here: https://gpt-index.readthedocs.io/en/latest/examples/vector_stores/MongoDBAtlasVectorSearch.html Large Language Models (LLMs) like ChatGPT have revolutionized the way users can get answers to their questions.
Build a ChatGPT with your Private Data using LlamaIndex and MongoDB
Co-authors: Prakul Agarwal - Senior Product Manager, Machine Learning at MongoDB Jerry Liu - co-founder at LlamaIndex Large Language Models (LLMs) like ChatGPT have revolutionized the way users can get answers to their questions. However, the "knowledge" of LLMs is restricted by what they were trained on, which for ChatGPT means publicly available information on the internet till September 2021.
Using LLM's for Retrieval and Reranking
This blog post outlines some of the core abstractions we have created in LlamaIndex around LLM-powered retrieval and reranking, which helps to create enhancements to document retrieval beyond naive top-k embedding-based lookup. LLM-powered retrieval can return more relevant documents than embedding-based retrieval, with the tradeoff being much higher latency and cost.
Testing Anthropic Claude's 100k-token window on SEC 10-K Filings
Anthropic's 100K Context Window expansion, just released yesterday, has taken the AI community by storm. A 100k token limit is approximately 75k words (~3x GPT4-32k's context window, ~25x that of GPT-3.5/ChatGPT); this means that you can now fit 300 pages of text in a single inference call. One of the core use cases highlighted in the Anthropic blog is analyzing an SEC 10-K filing; the model is capable of ingesting the entire report, and producing answers to different questions.
A Distilled Summary (using LlamaIndex)
I had the pleasure of speaking with Sam Charrington on the TWIML AI podcast about LlamaIndex, and the episode was just released this past Monday (5/8/23). I thought it would be a fun experiment to distill some highlights from the podcast! And what better way to do this than using LlamaIndex itself (plus OpenAI Whisper for transcription)? 😉 I did the following: Ran the podcast through whisper.cpp Did some light manual cleaning of the text, and uploaded it as a Dropbox file.
A New Document Summary Index for LLM-powered QA Systems
In this blog post, we introduce a brand new LlamaIndex data structure: a Document Summary Index. We describe how it can help offer better retrieval performance compared to traditional semantic search, and also walk through an example. One of the core use cases of Large Language Models (LLMs) is question-answering over your own data.
Unifying LLM-powered QA Techniques with Routing Abstractions
A variety of techniques for LLM-based QA over your data have emerged: e.g. semantic search, hybrid search for fact-based lookup, retrieving entire documents for summarization tasks, and more. Each technique is typically optimized for different query use cases. We believe “router” abstractions can help unify these techniques under a single query interface.
Unifying LLM-powered QA Techniques with Routing Abstractions
A variety of techniques for LLM-based QA over your data have emerged: e.g. semantic search, hybrid search for fact-based lookup, retrieving entire documents for summarization tasks, and more. Each technique is typically optimized for different query use cases. We believe "router" abstractions can help unify these techniques under a single query interface.
LlamaIndex and Deep Lake for Financial Statement Analysis
(co-authored by Davit Buniatyan, co-founder/CEO of Activeloop, and Jerry Liu, co-founder/CEO of LlamaIndex) Financial professionals, investors, and hobbyists need efficient ways to access, analyze, and make sense of this data to make informed decisions. Naturally, this is why building chatbots for your data tools to analyze earnings calls, financial report PDFs, or SEC filings is all the rage now.
LlamaIndex 0.6.0: A New Query Interface Over your Data
The central mission of LlamaIndex is to provide an interface between Large Language Models (LLM’s), and your private, external data. Over the past few months, it has become one of the most popular open-source frameworks for LLM data augmentation (context-augmented generation), for a variety of use cases: question-answering, summarization, structured queries, and more.
A New Query Interface Over your Data
The central mission of LlamaIndex is to provide an interface between Large Language Models (LLM's), and your private, external data. Over the past few months, it has become one of the most popular open-source frameworks for LLM data augmentation (context-augmented generation), for a variety of use cases: question-answering, summarization, structured queries, and more.
A Trace Mineral Injection before Joining and Lambing Increases Marking Percentages and Lamb Weights on Diverse Farms in Victoria, Australia
Open AccessArticle by Paula A. Gonzalez-Rivas 1,2,*, Graham R.
Show More
loading
Actions
Is this you?
As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.Get in touch with Jerry
Contact Jerry, search articles and posts on X, monitor coverage, and track replies from one place.
Learn more about Muck Rack