Person

Vicki Cheung

Vicki Cheung is a software engineer who was a founding member of OpenAI and later co-founded Gantry, an AI-model observability company.[1]first-party sourceIntroducing OpenAIOpenAIOpenAI’s 2015 introduction lists Cheung among its founding members; Observer later identifies her as a co-founder and CTO of Gantry.CheckedOpen original source [2]independent sourceWhere Are OpenAI’s Founding Members Today?ObserverObserver describes Cheung as an OpenAI founding member and later co-founder and CTO of Gantry.CheckedOpen original source

Markdown

On this page

01

Core facts

Entity type
Person
Occupations
Software engineer and technology entrepreneur · engineer[3]independent sourceIn Conversation With Vicki Cheung: SRE, MLOps, and Lessons LearnedHashiCorpIn a HashiCorp interview, Cheung discusses her software-engineering work across Duolingo, OpenAI and Lyft and the machine-learning infrastructure startup she was then developing.CheckedOpen original source
Official profile or website
vickicheung.com[4]first-party sourceVicki Cheung — Personal WebsiteVicki CheungCheung’s personal site identifies her by name and documents her software-engineering work at Duolingo and TrueVault.CheckedOpen original source

02

Who is Vicki Cheung?[3]independent sourceIn Conversation With Vicki Cheung: SRE, MLOps, and Lessons LearnedHashiCorpCheung’s interview documents her education, early Duolingo work, entry into OpenAI before its formation and infrastructure role.CheckedOpen original source [2]independent sourceWhere Are OpenAI’s Founding Members Today?ObserverObserver documents Cheung’s OpenAI infrastructure tenure and subsequent co-founding of Gantry.CheckedOpen original source

Cheung began working with the Duolingo team while studying at Carnegie Mellon University, contributing to its website, Android app, API and speech-recognition grading. She joined OpenAI before its public launch and helped build infrastructure for its early research team; in an interview she recalled being one of the first two people working with Greg Brockman. She later worked at Lyft and co-founded Gantry, where she was identified as chief technology officer in 2024.

03

History

  1. On 29 August 2016, OpenAI published “Infrastructure for deep learning,” co-authored by Vicki Cheung, Jonas Schneider, Ilya Sutskever and Greg Brockman. The post shared the lab’s early infrastructure design and accompanied the release of its Kubernetes EC2 autoscaler.[5]first-party sourceInfrastructure for deep learningOpenAIThe dated OpenAI article lists its authors, including Cheung, and describes publication of the autoscaler.CheckedOpen original source

  2. Joined OpenAI before its public formation and helped build early research infrastructure[3]independent sourceIn Conversation With Vicki Cheung: SRE, MLOps, and Lessons LearnedHashiCorpCheung recounts that she connected with the OpenAI team before the organization was formed and helped establish its initial infrastructure with Greg Brockman.CheckedOpen original source

04

Education

Carnegie Mellon
Cheung studied computer science at Carnegie Mellon University. While there, she approached professor Luis von Ahn about research work and began contributing to the early Duolingo project.[3]independent sourceIn Conversation With Vicki Cheung: SRE, MLOps, and Lessons LearnedHashiCorpIn the interview, Cheung recounts studying at Carnegie Mellon and starting Duolingo work after approaching professor Luis von Ahn.CheckedOpen original source

05

Career and roles

Duolingo
Cheung began contributing to Duolingo as a Carnegie Mellon student intern before the project spun out as a company; she continued as an early engineer, working across speech-recognition grading, web, Android and API components.[3]independent sourceIn Conversation With Vicki Cheung: SRE, MLOps, and Lessons LearnedHashiCorpCheung describes joining the project as a student intern, continuing with the team as it spun out, and taking on speech-recognition and full-stack work.CheckedOpen original source
Gantry
After her OpenAI and Lyft infrastructure roles, Cheung co-founded Gantry with machine-learning researcher Josh Tobin. Gantry emerged from stealth in 2022 as a company focused on helping teams monitor, evaluate and improve machine-learning products in production; TechCrunch identified Cheung and Tobin as its co-founders.[6]independent sourceGantry launches out of stealth to help keep AI models freshTechCrunchTechCrunch’s launch report identifies Cheung and Tobin as co-founders and describes Gantry’s production machine-learning improvement focus; Cheung discussed the startup in her 2021 interview.CheckedOpen original source
Lyft
At Lyft, Cheung led the Compute team responsible for building out a Kubernetes platform. In a 2019 interview, she described the work as a staged migration: the team began with batch and machine-learning workloads, validated the platform, and then moved production workloads onto Kubernetes.[7]independent sourceEpisode 80 — Lyft and KubeCon NA 2019, with Vicki CheungKubernetes Podcast from GoogleCheung describes leading the compute team, beginning Lyft’s Kubernetes migration with batch and machine-learning workloads, and reaching production use.CheckedOpen original source
OpenAI
Cheung joined OpenAI before its public launch and was part of its founding engineering group, leading infrastructure work. In interviews she described arriving before the organization had an office or established engineering systems and working with Greg Brockman to build what the research team needed.[7]independent sourceEpisode 80 — Lyft and KubeCon NA 2019, with Vicki CheungKubernetes Podcast from GoogleCheung recounts being OpenAI’s first engineer and building infrastructure before researchers arrived; the 2015 OpenAI introduction separately lists her in the founding group.CheckedOpen original source

06

Contributions and work

Duolingo Incubator
At Duolingo, Cheung led development of the Incubator, a web platform for creating language courses with contributions from volunteers and staff. Her description says the platform also maintained course data used by lesson-generation and answer-grading systems and exposed APIs for other internal teams.[4]first-party sourceVicki Cheung — Personal WebsiteVicki CheungCheung’s personal biography says she led the Incubator project and describes its course-authoring, lesson-data and internal API functions.CheckedOpen original source
Duolingo Speech Recognition
Her early Duolingo work included a speech-recognition grading engine for pronunciation exercises. Cheung’s account describes prototyping the engine and turning it into a scalable grading service; her own site also lists the speech-recognition grading engine among her Duolingo work.[3]independent sourceIn Conversation With Vicki Cheung: SRE, MLOps, and Lessons LearnedHashiCorpCheung recounts being assigned the speech-recognition engine at Duolingo; her personal site independently identifies the grading engine as part of her work.CheckedOpen original source
Gantry Ml Observability
Gantry’s launch coverage described a platform connecting application, labeling and storage data to help machine-learning teams inspect model behavior and decide when to retrain. This is a dated description of the company’s 2022 product direction, not a claim about its current availability.[6]independent sourceGantry launches out of stealth to help keep AI models freshTechCrunchTechCrunch reported Gantry’s launch from stealth and described its monitoring, data analysis and model-retraining use cases.CheckedOpen original source
Lyft Compute
Cheung’s Lyft infrastructure work focused on compute and Kubernetes adoption. She described the initial platform as a way to give service teams a consistent base while Lyft operated hundreds of microservices, rather than moving every production service at once.[7]independent sourceEpisode 80 — Lyft and KubeCon NA 2019, with Vicki CheungKubernetes Podcast from GoogleIn the interview, Cheung explains Lyft’s scale and why it began the Kubernetes transition with batch workloads before long-running services.CheckedOpen original source
OpenAI Deep Learning Infrastructure
Cheung co-authored OpenAI’s 2016 technical post on infrastructure for deep-learning research. The post explains the group’s use of Kubernetes to schedule jobs across physical and AWS machines, tools for experiment monitoring, and a batch-optimized autoscaler; it framed infrastructure as a way to improve researchers’ iteration and compute workflows.[5]first-party sourceInfrastructure for deep learningOpenAIThe August 2016 OpenAI post names Cheung as a co-author and documents the research infrastructure and orchestration systems described.CheckedOpen original source
OpenAI Kubernetes Autoscaler
As part of that infrastructure work, OpenAI released kubernetes-ec2-autoscaler, a Kubernetes node-scaling manager designed for batch workloads. The post explains that it adjusted AWS Auto Scaling groups in response to Kubernetes capacity needs, including memory/GPU constraints and regional capacity limits.[5]first-party sourceInfrastructure for deep learningOpenAIThe co-authored post describes the autoscaler’s batch focus, its interaction with AWS Auto Scaling groups, and handling of resource and region constraints.CheckedOpen original source

08

Sources

Reference numbers stay stable across this page and each source appears once, ordered by first visible use.

  1. Introducing OpenAI

    OpenAI · first-party source

  2. Vicki Cheung — Personal Website

    Vicki Cheung · first-party source

  3. Infrastructure for deep learning

    OpenAI · first-party source

  4. Episode 80 — Lyft and KubeCon NA 2019, with Vicki Cheung

    Kubernetes Podcast from Google · independent source

  5. OpenAI (Q21708200)

    Wikidata · structured public data source

Found something that needs review?

Suggest a correction