Builder

Jason Liu

Also known as jxnl, Instructor

Jason Liu is a Developer Experience Engineer on OpenAI’s Codex team. He created Instructor, a library that helps language models return structured data with validation. He studied mathematics at Waterloo, worked in data science at Facebook, and worked in ML engineering at Stitch Fix from 2018 to 2023. He then ran an independent consulting practice and courses on retrieval, agents, and evaluation. Those services have closed, while the course material remains available. His public writing explains how to build and improve AI systems through observed results. His open course material continues to explain retrieval systems, how to evaluate them, and how to use observed failures to guide changes.

Now at
OpenAI
Works
198
Citations
4,200
h-index
33

Publication numbers from OpenAlex, as of 14 October 2026.

Photo of Jason Liu

Timeline

  1. 2012–2017

    Studying mathematics and statistics

    Liu earned a mathematics degree at the University of Waterloo, with a focus on computational mathematics and statistics. He also published research using social media data.

  2. 2017–2023

    Working on machine learning systems

    Liu worked in data science at Facebook in 2017 and in machine learning at Stitch Fix from 2018 to 2023. His work involved embedding systems and internal software.

  3. 2023 onward

    Independent AI work

    Liu started 567 Studios in 2023. Liu earlier ran a solo consulting practice helping startups build AI products. The work covered retrieval, context, and evaluation. Consulting, expert calls, and course enrolments have closed.

  4. Public open source work

    Creating Instructor

    Instructor lets developers request structured results from language models with typed schemas and validation. Its public repository contains examples and integrations for model providers. The tool makes a practical application need explicit: a developer often needs fields that software can process and check.

  5. Public teaching work

    Improving retrieval systems

    Liu ran training on retrieval-augmented generation and AI agents. The course material remains an open resource. It connects his consulting work with reusable explanations of how to test a system, find problems, and make changes based on observed results.

  6. Current public role

    Developer experience for Codex

    Liu works on OpenAI’s Codex team. The site continues to collect technical notes and earlier course material.

Sources and credits

Photo: Jason Liu / GitHub profile. License: Not stated by source. Source: github.com. Cropped to a square and resized. Ask us to remove or replace this photo.

  1. Jason Liu: Résumé jxnl.co. Checked 11 October 2026.
  2. Instructor repository github.com. Checked 11 October 2026.
  3. Jason Liu: Introductions jxnl.co. Checked 11 October 2026.
  4. Jason Liu: Availability jxnl.co. Checked 11 October 2026.
  5. Systematically Improving RAG github.com. Checked 11 October 2026.
  6. Jason Liu: GitHub github.com. Checked 11 October 2026.