Jerry Liu

Jerry Liu is co-founder and CEO of LlamaIndex, the open-source framework and platform for connecting language models to private data through retrieval, indexing, and agents. He started the project as GPT Index in late 2022, then built a company around the idea that LLMs only become useful at work when they can see your documents, APIs, and databases.

From personal project to GPT Index

On Practical AI he recalled that the early name was casual: "It used to be called GPT Index, and I kind of made up that name because it sounded roughly relevant to what I was building at the time." The first experiments were personal hacks against OpenAI APIs that had been available for about a year. He released an open-source library around a "tree index" that used the model itself to organize and traverse information — before embeddings-and-RAG had hardened into best practice.

Making LLMs stateful around your data

As usage grew, the "index" metaphor got more concrete. On Practical AI he framed the mission as making it "really easy and powerful and fast and cheap to connect your language models with your own private data," wrapping a mostly stateless model call in a stateful service around your sources. On the dbt Labs podcast he distinguished in-context learning pipelines from fine-tuning: most teams need a software system that shoves the right context into the prompt, not a custom model for every knowledge base.

Building LlamaIndex the company

With co-founder Simon Suo he turned the OSS project into LlamaIndex the company, adding cloud services for production RAG, evaluation, and agent workflows while keeping the open packages as the entry point. The public product line stayed aimed at builders who need connectors, chunking, retrieval quality, and observability — the unglamorous layers between a demo chat and a trusted internal assistant.

On retrieval as the workhorse

In talks he argues that LLMs are strong decision engines that still need organizational knowledge and context, and that retrieval-augmented generation is the practical path for most enterprises. Agents and workflows sit on top of that retrieval layer. The through-line is data-centric: index well, retrieve well, then let the model reason — rather than pretending a frontier chat box already knows your company.