Job description

Amazon's Artificial General Intelligence (AGI) organization builds frontier models and the AI agents on top of them, and every one of them depends on data we can trust. The Frontier AI (FAI) Assessment team owns the science of dataset quality — for the data that trains frontier models, and for the benchmarks that determine whether a model or an agent actually works. In this role you will build the automated systems that make quality assessment at scale. Large language model (LLM)-as-a-Judge is the starting point, and our goal is to develop an agentic system that plan its own audits, critiques its own judgments, and improves its own accuracy over time. You will design it, calibrate it against expert human judgment, and set the technical direction for how the organization measures data quality. Key job responsibilities - Design the assessment methodology for frontier AI assets - Build the automated assessment system, from LLM-as-a-Judge baselines to agents that plan their own audits and learn from their own errors - Oversee the expert audit program and raise the proficiency of the auditors who run it - Design the measurement methods that make quality findings defensible, including sampling strategy and error analysis - Communicate findings to the teams that create the data and to the teams that train models and build agents using it - Set technical direction for assessment science, mentor junior scientists, and present results to senior leadership - Publish research on assessment methodology at top-tier venues and contribute to patents A day in the life - Train AI agents to assess the quality of datasets and evaluation benchmarks, and document where they fall short - Diagnose their failures and improve the models, rubrics, and calibration behind them - Review the quality reports that auditors and agents produce, and decide whether the conclusions hold - Coach expert auditors and junior scientists, and move more of the manual audit work into automation - Lead research on self-improving assessment agents, from open problem to publication About the team FAI Assessment is part of Frontier AI Assets in AGI. We assess the quality of the datasets and benchmarks behind Amazon's frontier models and agents, and we define what good data means. Today that work relies on human-in-the-loop review by domain experts. Our aim is to automate it with self-improving agents that experts keep calibrated. We are a small team of scientists working closely with the data, modeling, and agent teams. - 5+ years of building machine learning models for business application experience - Experience programming in Java, C++, Python or related language - Experience with neural deep learning methods and machine learning - PhD in Computer Science, Engineering, Mathematics, Statistics, Economics, or a related field - Experience with Machine Learning and Large Language Model fundamentals, including architecture, training/inference lifecycles, and optimization of model execution, or experience building complex software systems that have been successfully delivered to customers - Experience in machine learning, data mining, information retrieval, statistics or natural language processing, or experience designing or architecting (design patterns, reliability and scaling) of new and existing systems - Working knowledge of the public benchmarks used to evaluate frontier AI models and agents, and of their limitations - Experience designing large-scale human data collection and improving the quality of the datasets - Experience with the data pipelines behind model post-training and alignment - Experience in developing recursive self-improving agents based on frontier LLMs - Publication record at top-tier machine learning venues such as ICML, ICLR, CVPR, NeurIPS, AAAI, EMNLP, Interspeech, ICASSP - Experience in an academic research or teaching position, such as professor, lecturer, or postdoctoral researcher Amazon is an equal opportunity employer and does not discriminate on the basis of protected veteran status, disability, or other legally protected status. Our inclusive culture empowers Amazonians to deliver the best results for our customers. If you have a disability and need a workplace accommodation or adjustment during the application and hiring process, including support for the interview or onboarding process, please visit amazon.jobs for more information. If the country/region you’re applying in isn’t listed, please contact your Recruiting Partner. The base salary range for this position is listed below. Your Amazon package will include sign-on payments and restricted stock units (RSUs). Final compensation will be determined based on factors including experience, qualifications, and location. Amazon also offers comprehensive benefits including health insurance (medical, dental, vision, prescription, Basic Life & AD&D insurance and option for Supplemental life plans, EAP, Mental Health Support, Medical Advice Line, Flexible Spending Accounts, Adoption and Surrogacy Reimbursement coverage), 401(k) matching, paid time off, and parental leave. Learn more about our benefits at amazon.jobs . USA, WA, BELLEVUE - 167,100.00 - 226,100.00 USD annually

Requirements

Must-haves

Key job responsibilities
Design the assessment methodology for frontier AI assets
Build the automated assessment system, from LLM-as-a-Judge baselines to agents that plan their own audits and learn from their own errors
Oversee the expert audit program and raise the proficiency of the auditors who run it
Design the measurement methods that make quality findings defensible, including sampling strategy and error analysis
Communicate findings to the teams that create the data and to the teams that train models and build agents using it
Set technical direction for assessment science, mentor junior scientists, and present results to senior leadership
Publish research on assessment methodology at top-tier venues and contribute to patents
A day in the life
Train AI agents to assess the quality of datasets and evaluation benchmarks, and document where they fall short
Diagnose their failures and improve the models, rubrics, and calibration behind them
Review the quality reports that auditors and agents produce, and decide whether the conclusions hold
Coach expert auditors and junior scientists, and move more of the manual audit work into automation
Lead research on self-improving assessment agents, from open problem to publication
About the team
5+ years of building machine learning models for business application experience
Experience programming in Java, C++, Python or related language
Experience with neural deep learning methods and machine learning
PhD in Computer Science, Engineering, Mathematics, Statistics, Economics, or a related field
Working knowledge of the public benchmarks used to evaluate frontier AI models and agents, and of their limitations
Experience designing large-scale human data collection and improving the quality of the datasets
Experience with the data pipelines behind model post-training and alignment
Experience in developing recursive self-improving agents based on frontier LLMs
Publication record at top-tier machine learning venues such as ICML, ICLR, CVPR, NeurIPS, AAAI, EMNLP, Interspeech, ICASSP

Nice-to-haves

None flagged.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aillmmachine learningjavapython

Platform: amazon
Effort: medium

Benefits

Benefits detected in the listing.

Health Insurance

Dental

Vision

401(K)

Paid Time Off

Parental Leave