Social
Conversational AI

A quick guide to Amazon’s 50-plus papers at EMNLP 2024

Large language models predominate, both as a research subject themselves and as tools for researching topics of particular interest to Amazon, such as speech, recommendations, and information retrieval.

By Staff writer
November 14, 2024
7 min read
Share

Large language models (LLMs) have come to dominate the field of natural-language processing, so it’s no surprise that they also dominate the research that Amazon scientists are presenting at this year’s Conference on Empirical Methods in Natural-Language Processing (EMNLP). LLM training is the topic with the greatest number of Amazon papers, followed closely by strategies for mitigating misinformation in LLMs’ outputs — including but not limited to hallucinations. At the same time, a number of papers apply LLMs to topics of traditional interest at Amazon, such as speech, recommender systems, and information retrieval. (Papers marked with asterisks were accepted to Findings of EMNLP.)

AI agents

MARCO: Multi-agent real-time chat orchestration
Anubhav Shrimal, Shervin Malmasi, Kriti Biswas, Swarnalatha Raghuraman, Anish Nediyanchath, Yi Zhang, Promod Yenigalla

Code generation

CodeFort: Robust training for code generation models
Yuhao Zhang, Shiqi Wang, Haifeng Qian, Zijian Wang, Mingyue Shang, Linbo Liu, Sanjay Krishna Gouda, Baishakhi Ray, Murali Krishna Ramanathan, Xiaofei Ma, Anoop Deoras

Socratic human feedback (SoHF): Expert steering strategies for LLM code generation
Subramanian Chidambaram, Erran Li, Min Bai, Xiaopeng LI, Kaixiang Lin, Xiong Zhou, Alex C. Williams

Structured object language modeling (SoLM): Native structured objects generation conforming to complex schemas with self-supervised denoising
Amir Tavanaei, Kee Kiat Koo, Hayreddin Ceker, Shaobai Jiang, Qi Li, Julien Han, Karim Bouyarmane

Contrastive decoding

Explaining and improving contrastive decoding by extrapolating the probabilities of a huge and hypothetical LM
Haw-Shiuan Chang, Nanyun Peng, Mohit Bansal, Anil Ramakrishna, Tagyoung Chung

Explaining and improving contrastive decoding by extrapolating the probabilities of a huge and hypothetical LM.png
Given a simple question with clues, contrastive decoding could have an “obvious blindness” (e.g., assigning higher probability to an uncommon answer, such as "invertebrate", than to the most obvious answer, "bees"). In contrast, the asymptotic probability decoding proposed in "Explaining and improving contrastive decoding by extrapolating the probabilities of a huge and hypothetical LM" correctly assigns the highest probability to "bees" by leveraging the probabilities from multiple LMs of different sizes.

Data integration

ASTRA: Automatic schema matching using machine translation
Tarang Chugh, Deepak Zambre

Learning from natural language explanations for generalizable entity matching
Somin Wadhwa, Adit Krishnan, Runhui Wang, Byron C. Wallace, Chris (Luyang) Kong

Pretraining and finetuning language models on geospatial networks for accurate address matching
Saket Maheshwary, Arpan Paul, Saurabh Sohoney

Retrieval augmented spelling correction for e-commerce applications
Xuan Guo, Rohit Patki, Dante Everaert, Christopher Potts

Dataset distillation

Textual dataset distillation via language model embedding
Yefan Tao, Chris (Luyang) Kong, Andrey Kan, Laurent Callot

Textual dataset distillation via language model embedding: DaLLME.png
The DaLLME framework proposed in "Textual dataset distillation via language model embedding" begins by using a language model to transform raw textual data into embedding vectors. A set of distilled vectors is then derived in the embedding space, through a process designed to encapsulate maximum informational content. Finally, the vec2text model translates these distilled vectors back into textual form.

Document understanding

DocKD: Knowledge distillation from LLMs for open-world document understanding models
Sungnyun Kim, Haofu Liao, Srikar Appalaraju, Peng Tang, Zhuowen Tu, Ravi Kumar Satzoda, R. Manmatha, Vijay Mahadevan, Stefano Soatto

Information retrieval

Evaluating D-MERIT of partial-annotation on information retrieval
Royi Rassin, Yaron Fairstein, Oren Kalinsky, Guy Kushilevitz, Nachshon Cohen, Alexander Libov, Yoav Goldberg

Identifying high consideration e-commerce search queries
Zhiyu Chen, Jason Choi, Besnik Fetahu, Shervin Malmasi

Learning when to retrieve, what to rewrite, and how to respond in conversational QA*
Nirmal Roy, Leonardo Ribeiro, Rexhina Blloshmi, Kevin Small

Natural-language understanding

Intent detection in the age of LLMs
Gaurav Arora, Shreya Jain, Srujana Merugu

Intent detection in the age of LLMs.png
"Intent detection in the age of LLMs" proposes a methodology for adaptive in-context learning and chain-of-thought-based intent detection using LLMs.

Predicting entity salience in extremely short documents
Ben Bullough, Harrison Lundberg, Chen Hu, Weihang Xiao

LLM evaluation

AXCEL: Automated eXplainable consistency evaluation using LLMs*
P Aditya Sreekar, Sahil Verma, Suransh Chopra, Sarik Ghazarian, Abhishek Persad, Narayanan Sadagopan

Precise model benchmarking with only a few observations
Riccardo Fogliato, Pratik Patil, Nil-Jana Akpinar, Mathew Monfort

LLM fine tuning

AdaZeta: Adaptive zeroth-order tensor-train adaption for memory-efficient large language models fine-tuning
Yifan Yang, Kai Zhen, Ershad Banijamali, Thanasis Mouchtaris, Zheng Zhang

RoseLoRA: Row and column-wise sparse low-rank adaptation of pre-trained language model for knowledge editing and fine-tuning
Haoyu Wang, Tianci Liu, Ruirui Li, Monica Cheng, Tuo Zhao, Jing Gao

RoseLoRA.png
The row- and column-wise sparse low-rank adaptation (RoseLoRA) framework proposed in "RoseLoRA: Row and column-wise sparse low-rank adaptation of pre-trained language model for knowledge editing and fine-tuning".

LLMs for speech

Speechworthy instruction-tuned language models
Hyundong Cho, Nicolaas Jedema, Leonardo Ribeiro, Karishma Sharma, Pedro Szekely, Alessandro Moschitti, Ruben Janssen, Jonathan May

LLM misinformation mitigation

ECON: On the detection and resolution of evidence conflicts
Cheng Jiayang, Chunkit Chan, Qianqian Zhuang, Lin Qiu, Tianhang Zhang, Tengxiao Liu, Yangqiu Song, Yue Zhang, Pengfei Liu, Zheng Zhang

Generative subgraph retrieval for knowledge graph–grounded dialog generation
Jinyoung Park, Minseok Joo, Joo-Kyung Kim, Hyunwoo J. Kim

HalluMeasure: Fine-grained hallucination measurement using chain-of-thought reasoning
Shayan Ali Akbar, Md Mosharaf Hossain, Tess Wood, Si-Chi Chin, Erica Salinas, Victor Alvarez, Erwin Cornejo

Knowledge-centric hallucination detection
Xiangkun Hu, Dongyu Ru, Lin Qiu, Qipeng Guo, Tianhang Zhang, Yang Xu, Yun Luo, Pengfei Liu, Zheng Zhang, Yue Zhang

LLM reasoning

Auto-evolve: Enhancing large language model’s performance via self-reasoning framework*
Krishna Aswani, Alex Lu, Pranav Patankar, Priya Dhalwani, Iris Tan, Jayant Ganeshmohan, Simon Lacasse

LLM self-correction

LLM self-correction with DeCRIM: Decompose, critique, and refine for enhanced following of instructions with multiple constraints
Thomas Palmeira Ferraz, Kartik Mehta, Yu-Hsiang Lin, Haw-Shiuan Chang, Shereen Oraby, Sijia Liu, Vivek Subramanian, Tagyoung Chung, Mohit Bansal, Nanyun Peng

DeCRIM.png
In the DeCRIM pipeline proposed in "LLM self-correction with DeCRIM: Decompose, critique, and refine for enhanced following of instructions with multiple constraints", an LLM first generates a response to a user request. The Decomposer then breaks down the request into granular constraints, and the Critic model gives feedback on whether the response meets those constraints. If it does, the response is output; if not, the LLM uses the feedback to refine the response.

LLM training

Dancing in chains: Reconciling instruction following and faithfulness in language models
Zhengxuan Wu, Yuhao Zhang, Peng Qi, Yumo Xu, Rujun Han, Yian Zhang, Jifan Chen, Bonan Min, Zhiheng Huang

DEM: Distribution edited model for training with mixed data distributions
Dhananjay Ram, Aditya Rawal, Momchil Hardalov, Nikolaos Pappas, Sheng Zha

DEM: Distribution Edited Model for Training with Mixed Data Distributions
The distribution-edited model (ΘD) described in "DEM: Distribution edited model for training with mixed data distributions" results from fine-tuning a pretrained model (Θ) on n individual data distributions (Di) and combining the resulting models with basic element-wise vector operations. Here, the extracted distribution vectors (∆ΘDi ) are multiplied by weight coefficients, and the weighted sum is added to the base model.

Evolutionary contrastive distillation for language model alignment
Julian Katz-Samuels, Zheng Li, Hyokun Yun, Priyanka Nigam, Yi Xu, Vaclav Petricek, Bing Yin, Trishul Chilimbi

Hop, skip, jump to convergence: Dynamics of learning rate transitions for improved training of large language models
Shreyas Subramanian, Vignesh Ganapathiraman, Corey Barrett

Learning from relevant subgoals in successful dialogs using iterative training for task-oriented dialog systems
Magdalena Kaiser, Patrick Ernst, Gyuri Szarvas

Quality matters: Evaluating synthetic data for tool-using LLMs
Shadi Iskander, Nachshon Cohen, Zohar Karnin, Ori Shapira, Sofia Tolmach

Query autocompletion

AmazonQAC: A large-scale, naturalistic query autocomplete dataset
Dante Everaert, Rohit Patki, Tianqi Zheng, Christopher Potts

DiAL: Diversity aware listwise ranking for query auto-complete
Sonali Singh, Sachin Farfade, Prakash Mandayam Comar

Question answering

RAG-QA arena: Evaluating domain robustness for long-form retrieval-augmented question answering
Rujun Han, Yuhao Zhang, Peng Qi, Yumo Xu, Jenyuan Wang, Lan Liu, William Yang Wang, Bonan Min, Vittorio Castelli

Retrieving contextual information for long-form question answering using weak supervision
Philipp Christmann, Svitlana Vakulenko, Ionut Teodor Sorodoc, Bill Byrne, Adrià de Gispert

Recommender systems

Efficient pointwise-pairwise learning-to-rank for news recommendation
Nithish Kannen Senthilkumar, Yao Ma, Gerrit van den Burg, Jean Baptiste Faddoul

Efficient pointwise-pairwise learning-to-rank for news recommendation.png
An illustration of the GLIMPSE framework proposed in "Efficient pointwise-pairwise learning-to-rank for news recommendation". GLIMPSE adopts a multitask approach in which a pretrained language model is fine-tuned on both the relevance prediction task and the pairwise-preference task. During inference, the relevance predictions are used to produce an initial pointwise ranking, which is subsequently improved by one or more right-to-left (RTL) passes using pairwise comparisons.

PEARL: Preference extraction with exemplar augmentation and retrieval with LLM agents
Vijit Malik, Akshay Jagatap, Vinayak Puranik, Anirban Majumder

Sequential LLM framework for fashion recommendation
Han Liu, Xianfeng Tang, Tianlang Chen, Jiapeng Liu, Indu Indu, Henry Peng Zou, Peng Dai, Roberto Fernandez Galan, Mike Porter, Dongmei Jia, Ning Zhang, Lian Xiong

Responsible AI

Attribute controlled fine-tuning for large language models: A case study on detoxification
Tao Meng, Ninareh Mehrabi, Palash Goyal, Anil Ramakrishna, Aram Galstyan, Richard Zemel, Kai-Wei Chang, Rahul Gupta, Charith Peris

FLIRT: Feedback loop in-context red teaming
Ninareh Mehrabi, Palash Goyal, Christophe Dupuy, Qian Hu, Shalini Ghosh, Richard Zemel, Kai-Wei Chang, Aram Galstyan, Rahul Gupta

Order of magnitude speedups for LLM membership inference
Rongting Zhang, Martin Bertran Lopez, Aaron Roth

Synthetic data generation

CorrSynth: A correlated sampling method for diverse dataset generation from LLMs
Suhas Kowshik, Abhishek Divekar, Vijit Malik

A Correlated Sampling Method for Diverse Dataset Generation from LLMs
"CorrSynth: A correlated sampling method for diverse dataset generation from LLMs" introduces a sampling method that uses anti-correlation between examples rather than few-shot generation.

DATA ADVISOR: Dynamic data curation for safety alignment of large language models
Fei Wang, Ninareh Mehrabi, Palash Goyal, Rahul Gupta, Kai-Wei Chang, Aram Galstyan

Evaluating differentially private synthetic data generation in high-stakes domains
Krithika Ramesh, Nupoor Gandhi, Pulkit Madaan, Lisa Bauer, Charith Peris, Anjalie Field

SYNTHESIZRR: Generating diverse datasets with retrieval augmentation
Abhishek Divekar, Greg Durrett

Abstract depiction of the SYNTHESIZRR procedure
Abstract depiction of the procedure proposed in "SYNTHESIZRR: Generating diverse datasets with retrieval augmentation". The content sourcing stage retrieves K unique documents {r1,...,rK} from a large corpus for each in-context covariate xICL. The task-inversion stage uses a parameterized context refinement prompt, Pτ, which takes parameters Rinv (inversion instruction), rk (a retrieved document), and V(yICL) (the verbalized target label). A generalist teacher LLM autoregressively generates a synthetic covariate. Each in-context example thus produces K unique synthetic examples {x̃1,..., x̃K}, which we include in the dataset with target yICL.

Text classification

Distance-aware calibration for pre-trained language models*
Alberto Gasparin, Gianluca Detommaso

Performance-guided LLM knowledge distillation for efficient text classification at scale

Flavio Di Palo, Prateek Singhi, Bilal Fadlallah

Prompt-tuned muti-task taxonomic transformer (PTMTTaxoFormer)
Rajashekar Vasantha, Nhan Nguyen, Yue Zhang

Text summarization

Salient information prompting to steer content in prompt-based abstractive summarization
Lei Xu, Asad Karim, Saket Dingliwal, Aparna Elangovan

Research areas
Tags
About the Author
Staff writer

Related content

Work with us

See more jobs See more jobs
Applied Scientist, Structural Design Integrity, Hardware - Product Integrity
CN, 31, Shanghai
As an Applied Scientist, you will be responsible for bringing new product designs through to manufacturing. You will work closely with multi-disciplinary groups including Product Design, Industrial Design, Hardware Engineering, and Operations, to drive key aspects of engineering of consumer electronics products. In this role, you will use expertise in physical sciences, theoretical, numerical or empirical techniques to create scalable models representing response of physical systems or devices, including: * Applying domain scientific expertise towards developing innovative analysis and tests to study viability of new materials, designs or processes * Working closely with engineering teams to drive validation, optimization and implementation of hardware design or software algorithmic solutions to improve product and customer risks * Establishing scalable, efficient, automated processes to handle large scale design and data analysis * Conducting research into use conditions, materials and analysis techniques * Tracking general business activity including device health in field and providing clear, compelling reports to management on a regular basis * Developing, implementing guidelines to continually optimize design processes * Using simulation tools like LS-DYNA, and Abaqus for analysis and optimization of product design * Using of programming languages like Python and Matlab for analytical/statistical analyses and automation * Demonstrating strong understanding across multiple physical science domains, e.g. structural, thermal, fluid dynamics, and materials * Developing, analyzing and testing structural solutions from concept design, feature development, product architecture, through system validation * Supporting product development and optimization through application of analysis and testing of complex electronic assemblies using advanced simulation and experimentation tools and techniques
Applied Scientist, Personalization, Personalization
IL, Haifa
We are seeking an Applied Scientist to help build Amazon’s next-generation customer memory and personalization systems. Are you interested in building systems that move beyond reacting to customer behavior, to actually understanding and remembering it over time? Our team is building Amazon’s customer memory layer – a system that extracts, curates, and reasons over customer knowledge to power next-generation personalization. This includes transforming noisy, unstructured signals into durable, high-quality representations of customer preferences, intents, and life events, and using them in real time to improve customer experiences. We are part of Amazon’s Personalization organization, a high-performing group that leverages large-scale machine learning, generative AI, and distributed systems to deliver highly relevant customer experiences. We tackle challenging problems at the intersection of information extraction, knowledge representation, LLM reasoning, and recommendation systems. Our systems operate under real-world constraints of scale, latency, and quality, requiring careful tradeoffs between precision, recall, and responsiveness. This team plays a central role in defining how Amazon understands its customers, and how that understanding is applied across the shopping experience. As an Applied Scientist, you will design and build ML and LLM-powered solutions for Amazon's customer memory and personalization systems. You will work on how customer knowledge is extracted, validated, and applied in production systems. You will own the end-to-end delivery of ML solutions, from problem formulation and modeling to offline and online experimentation, and production deployment at scale. You will deliver high-quality, scalable systems that power customer-facing experiences. You will drive work across areas such as fact extraction, memory quality and lifecycle, temporal reasoning, and grounded personalization, while navigating tradeoffs between quality, latency, and coverage. You will collaborate closely with engineering and product teams to translate research into measurable customer impact. Please visit https://www.amazon.science for more information.
Sr Data Scientist, Amazon - Vertical Ads
US, WA, Seattle
Are you passionate about solving big problems from ground-up? Do you enjoy building new state-of-the-art products at internet scale? Come lead the innovation in this startup team, vertical ad products. This is a green field problem without a known answer or a pattern to follow. We have ambitious vision to simplify full funnel advertising solutions, at scale, with specialized agentic AI-powered models and diversify the demand to strategic verticals including finserv, autos, locals.. etc. We are seeking an experienced Sr Data Scientist to drive innovation in our Ads Foundational Model. In this individual contributor role, you will apply advanced machine learning techniques to improve advertiser performance and customer experience. Key job responsibilities As a Data Scientist on this team, you will: 1. Develop and drive the science strategy for Ads Foundational Model (Ads-FM), aligning it with the program's objectives and overall business goals. 2. Identify high-impact opportunities within Ads-FM program and lead the ideation, planning, and execution of science initiatives to address them. 3. Build and deploy machine learning models using computer vision, natural language processing, and deep learning to evaluate and enhance ad effectiveness. 4. Develop algorithms that extract meaningful signals from image, video, and audio content to predict and improve customer engagement 5. Leverage Amazon's extensive data repository to create predictive models that generate actionable recommendations for more compelling ad creative 6. Collaborate with business leaders and cross-functional teams to implement ML-powered solutions 7. Contribute to the ML roadmap for the Ads-FM program through innovation and research.
Director, Economist - Head of Experimentation & Value, Devices Science Optimization (DSO)
US, WA, Seattle
You will build and lead the economics research agenda for measurement, experimentation, and value attribution for Amazon's Devices & Services organization. Your team is the "truth layer" of the Intelligence Core — the shared economics and causal inference capability that serves all Devices product lines, marketing pods, and Finance leadership with causal evidence of what Devices are worth and whether our investments are working. This is not a traditional analytics or measurement role. You will own an active research program in experimentation design — identifying and executing the causal studies that produce the causal inputs for pricing decisions, marketing optimization, and portfolio strategy. Your outputs provide the causal evidence base that L8 peers and senior leadership consume to make billions of dollars in investment decisions across the D&S portfolio. You will also own the economic models that validate and drive execution across the full surface area of marketing spend for devices and services. Key job responsibilities Economic Value: • Downstream value attribution for all Devices product lines — Impact on Prime, subscription lift, consumer spending, advertising value • Alexa+ value isolation and cross-PL attribution • Causal frameworks connecting device sales to Prime acquisition, subscription retention, and ecosystem engagement Marketing Science & Measurement: • Build the marketing science function from scratch • Incrementality measurement for marketing spend across all channels • Attribution methodology, measurement standards, and cross-pod governance • Marketing ROI frameworks for use by category marketers • CCM certification methodology and scenario planning models for optimal investment allocation Experimentation: • Owning the estimation methodology, identification strategies, data inputs/outputs, and refresh cadence • You will build this team's analytics function with AI at its core from day one • Experimentation governance — managing interference across teams, setting standards for causal validity • Evaluation framework for AI agents and autonomous optimization systems
Quality Assurance Manager, Special Projects
US, WA, Seattle
Innovators wanted! Are you an entrepreneur? A builder? A dreamer? This role is part of an Amazon Special Projects team that takes the company’s Think Big leadership principle to the extreme. We focus on creating entirely new products and services with a goal of positively impacting the lives of our customers. No industries or subject areas are out of bounds. If you’re interested in innovating at scale to address big challenges in the world, this is the team for you. Here at Amazon, we embrace our differences. We are committed to furthering our culture of inclusion. We have thirteen employee-led affinity groups, reaching 40,000 employees in over 190 chapters globally. We are constantly learning through programs that are local, regional, and global. Amazon’s culture of inclusion is reinforced within our 16 Leadership Principles, which remind team members to seek diverse perspectives, learn and be curious, and earn trust. Our team highly values work-life balance, mentorship and career growth. We believe striking the right balance between your personal and professional life is critical to life-long happiness and fulfillment. We care about your career growth and strive to assign projects and offer training that will challenge you to become your best.
Manager Research Science, Last Mile Science and Analytics
IN, KA, Bengaluru
Have you ever wondered how that Amazon box with the smile arrives so quickly, where it came from, and how much it cost Amazon to deliver? The WW Amazon Logistics, Business Analytics team manages the delivery of tens of millions of products every week to Amazon's customers, achieving on-time delivery in a cost-effective manner. We are seeking an enthusiastic, customer-obsessed Manager Research Science with strong analytical skills to join our team. This role is crucial in optimizing Amazon's vast delivery network and will have significant impact on the customer experience, particularly in the final phase of delivery. As a Manager Research Science, you will: 1. Address business challenges through building compelling cases and using data to influence change across the organization 2. Develop input and assumptions based on preexisting models to estimate costs and savings opportunities associated with varying levels of network growth and operations 3. Create metrics to measure business performance, identify root causes and trends, and prescribe action plans 4. Manage multiple high-impact projects simultaneously 5. Work with technology teams and product managers to develop new tools and systems supporting business growth 6. Communicate with and support various internal stakeholders and external audiences 7. Implement scheduling solutions, improve metrics, and develop scalable processes and tools The ideal candidate will have: - Extensive experience in operations research and data-driven decision making - Strong analytical and problem-solving skills - Robust program management and research science skills - Ability to work with a team and make independent decisions in ambiguous environments - Customer-obsessed mindset with a focus on improving the Amazon delivery experience This role offers the autonomy to think strategically and make data-driven decisions from day one. Join us in shaping the future of e-commerce delivery and addressing the core challenges in our world-class operations space! Key job responsibilities 1. Advanced Modeling and Algorithm Development: - Design and implement sophisticated machine learning models for logistics optimization - Develop complex time series forecasting algorithms for demand prediction and resource allocation 2. AI and Machine Learning Integration: - Architect and deploy AI-powered systems to enhance decision-making in logistics operations - Implement deep learning techniques for image recognition in package sorting and handling - Develop reinforcement learning algorithms for adaptive scheduling and resource management 3. Big Data Analytics and Processing: - Design and implement distributed computing solutions for processing massive logistics datasets - Utilize cloud computing platforms (e.g., AWS) for scalable data processing and analysis 4. AI-Driven Workflow Optimization: - Design and implement AI agents for autonomous decision-making in logistics processes - Create machine learning models for customer behavior analysis and personalized delivery options 5. Software Development and System Architecture: - Write efficient, scalable code in languages such as Python, Java, or C++ - Develop and maintain complex software systems for logistics optimization - Stay at the forefront of AI and ML research - Publish research findings in top-tier conferences and journals About the team We are Amazon's Last Mile Science and Analytics team, dedicated to improving e-commerce delivery. We work to optimize our vast network, forecast demand using machine learning, and enhance route efficiency. Our efforts focus on developing innovative delivery methods, applying AI to solve complex problems, and conducting geospatial analysis. We create simulations to refine processes and plan capacity effectively. Operating globally, we strive to develop adaptable solutions for diverse markets. We aim to advance logistics science, continually improving speed, efficiency, and customer satisfaction, in support of Amazon's mission to be Earth's most customer-centric company.
Applied Scientist - LLM, Alexa Conversational Modelling Intelligence
DE, BE, Berlin
As an Applied Scientist II in the Alexa Conversational Modelling Intelligence team within Alexa AI, you will drive model post-training for Large Language Models that power Alexa+. You'll adopt and adapt state-of-the-art techniques — including supervised fine-tuning, RLHF, and preference optimization — running rigorous experiments and translating findings into production-ready solutions that directly improve the customer experience for millions of users worldwide. You will own the full model development cycle from data curation through training, evaluation, and deployment. Your day-to-day will involve developing evaluation methods and metrics, diagnosing model defects, and iterating on recipes to move concrete quality and efficiency benchmarks. You'll write clean, reproducible code, contribute to shared tooling, and collaborate closely with scientists and engineers to bring models from experimentation to scale. You are technically curious, experiment-driven, and motivated by real customer impact. You will also advance the state of the art by publishing at top-tier NLP/ML conferences (ACL, EMNLP, NeurIPS, ICML, ICLR) — contributing to the broader research community while grounding your work in measurable outcomes. Key job responsibilities As an Applied Scientist II in the Alexa Conversational Modelling Intelligence team, you will own the end-to-end model development lifecycle for LLMs that power Alexa+. You'll design and execute training recipes — including supervised fine-tuning, reinforcement learning from human feedback, and preference optimization — iterating rapidly on data, hyperparameters, and architectures to move quality and efficiency metrics. Your work will directly shape how millions of customers interact with Alexa daily. You will build robust evaluation frameworks to measure model performance, diagnose failure modes, and quantify improvements. This includes developing benchmarks, implementing LLM-as-a-judge pipelines, and conducting rigorous defect analysis to identify where models fall short and why. You'll translate these insights into targeted improvements that close gaps in conversational quality, safety, and fluency. You will collaborate closely with research scientists and engineers to bring models from experimentation to production at scale. You'll contribute to shared tooling and infrastructure, write clean and reproducible code, and document your methods so the team can build on your work. You are also expected to advance the state of the art by publishing findings at top-tier NLP/ML venues (ACL, EMNLP, NeurIPS, ICML, ICLR), ensuring your research drives both customer impact and scientific contribution. A day in the life As an Applied Scientist II, your day will involve launching and monitoring training runs, analyzing experiment results, and iterating on model recipes based on evaluation data. You'll participate in science reviews with fellow researchers, sync with engineering partners on deployment readiness, and deep-dive into model outputs to understand behavioral patterns. You'll balance hands-on experimentation with collaborative problem-solving — working across the Alexa AI organization to align model improvements with customer-facing goals and product priorities. About the team The Alexa Conversational Modelling Intelligence team builds industry-leading LLM-based conversational technologies that customers love. Our mission is to push the envelope in LLMs for Alexa to deliver the best-possible customer experience. As an Applied Scientist, you'll contribute directly to that mission through model development and experimentation.
Manager III, Economist - AMZ9803624
US, CA, Sunnyvale
MULTIPLE POSITIONS AVAILABLE Employer: AMAZON.COM SERVICES LLC Offered Position: Manager III, Economist Job Location: Sunnyvale, California Job Number: AMZ9803624 Position Responsibilities: Independently manage a team of economists and/or scientists in developing strategic economic analyses and demand estimation models. Translate business questions into econometric methodologies and causal inference analyses. Communicate economic insights to non-technical audiences to guide strategic-level, high-impact business decisions. Scale economic models through cross-functional collaboration with engineering teams. Establish scientific quality standards and research priorities. Drive operational efficiency and research excellence across the team. 40 hours / week, 8:00am-5:00pm, Salary Range: $201,300/year to $272,400/year. Amazon is a total compensation company. Dependent on the position offered, equity, sign-on payments, and other forms of compensation may be provided as part of a total compensation package, in addition to a full range of medical, financial, and/or other benefits. For more information, visit: https://www.aboutamazon.com/workplace/employee-benefits. Amazon.com is an Equal Opportunity-Affirmative Action Employer – Minority / Female / Disability / Veteran / Gender Identity / Sexual Orientation.#0000
Economist, Prime Video
US, WA, Seattle
Ever wish you could use your quantitative and critical thinking skills to influence business decisions? Economists at Amazon partner closely with senior management, business stakeholders, scientist and engineers, and economist leadership to solve key business problems. As part of the Content Discovery and Experimentation Science team within Prime Video, you will leverage your expertise in causal inference and experimental design to make Prime Video the best-in-class digital video experience. Key job responsibilities - Build causal models and metrics that capture trade-off decisions when business and customer outcomes do not align - Partner with data scientists and product managers to integrate these metrics into Prime Video's experimentation tooling - Work with finance partners to ensure that the team's product metrics contribute to Prime Video's strategic business and financial objectives - Contribute to technical and business documents to communicate ideas and proposals to various audiences - Educate and advocate for best practices in experimentation and how to use it for decision-making
Senior Data Scientist, Enterprise Security Products
US, TX, Austin
What happens when you combine startup speed with Amazon-scale impact? You get this team. Amazon Enterprise Security Products is a newly launched group building intelligent, cloud-agnostic security tools using AI-first development practices. Here, you build AI and you build with AI at the same time. This role is a chance to define and lead the science strategy for the future of security tooling with a small, fast team that ships like a startup but deploys at Amazon scale. We're looking for a Senior Data Scientist who operates at the intersection of applied ML, agentic AI, and security; and who can set technical direction across ambiguous, undefined problem spaces. You won't just build models; you'll decide which problems are worth solving, architect the scientific approach for an entire product area, and raise the bar for how the team applies science. You'll partner with senior and principal engineers, applied scientists, security researchers, and PMs, and your judgment will shape roadmaps, not just deliverables. This is a role for someone who thrives in ambiguity, influences without authority, and turns "too ambitious" into shipped reality. Key job responsibilities - Set the science direction for a product area: Define the modeling strategy, scientific approach, and success metrics for entire categories of AI-first security capabilities, agentic systems, anomaly detection, threat classification, and automated response across multi-cloud environments. Decide where science can move the needle and where it can't. - Own the hardest, most ambiguous problems: Take on undefined, open-ended challenges where the path isn't clear, the data is messy or scarce, and the stakes are high. Frame the problem, choose the approach, and bring others along. - Build with AI to build AI and define how the team does it: Drive adoption of agentic coding tools, LLM-powered workflows, and experimental AI tooling across the science org. Establish the practices that multiply velocity for every scientist, not just yourself. - Architect agentic intelligence: Lead the design of models, embeddings, RAG pipelines, evaluation frameworks, and feedback loops that make multi-agent security systems smart, safe, and customer-ready at scale. Own the science architecture decisions others build on. - Drive technical strategy across teams: Influence roadmaps, dive deep with senior and principal scientists and engineers, and align cross-functional partners around a shared scientific vision. Your recommendations shape what the team invests in next. - Prototype, validate, and scale: Turn ambiguous hypotheses into prototypes in days, validate with real customer signal, and chart the path from prototype to production system that runs reliably at Amazon scale. - Communicate to influence at the executive level: Translate complex modeling results and scientific trade-offs into clear recommendations for engineers, product leaders, and senior executives. Drive organizational decisions with data and earn trust across the company. - Raise the bar and grow others: Mentor data scientists and applied scientists, lead technical and science reviews, and champion AI-first development practices. Shape the science culture and hiring bar of a fast-growing team from the ground floor. A day in the life No two days look the same on this fast-growing, AI-first team. You might start your morning setting direction in a roadmap review; making the call on which science investments will have the biggest customer impact and then dive into architecting an evaluation framework that the whole team will build on. Before lunch, you're pair-prompting with an agentic coding assistant to validate a new approach, then unblocking a teammate stuck on a thorny modeling problem. In the afternoon, you lead a design session with senior and principal scientists and engineers, then distill it into a crisp recommendation for senior leadership. You own ambiguous problems end to end, define how the team works, and see your decisions ripple across the product. This is where builders who want to lead with science come to do their best work. About the team Why AWS? Amazon Web Services (AWS) is the world’s most comprehensive and broadly adopted cloud platform. We pioneered cloud computing and never stopped innovating — that’s why customers from the most successful startups to Global 500 companies trust our robust suite of products and services to power their businesses. Inclusive Team Culture Here at AWS, it’s in our nature to learn and be curious. Our employee-led affinity groups foster a culture of inclusion that empower us to be proud of our differences. Ongoing events and learning experiences, including our Conversations on Race and Ethnicity (CORE) and AmazeCon conferences, inspire us to never stop embracing our uniqueness. Mentorship & Career Growth We’re continuously raising our performance bar as we strive to become Earth’s Best Employer. That’s why you’ll find endless knowledge-sharing, mentorship and other career-advancing resources here to help you develop into a better-rounded professional. Work/Life Balance We value work-life harmony. Achieving success at work should never come at the expense of sacrifices at home, which is why we strive for flexibility as part of our working culture. When we feel supported in the workplace and at home, there’s nothing we can’t achieve in the cloud. Hybrid Work We value innovation and recognize this sometimes requires uninterrupted time to focus on a build. We also value in-person collaboration and time spent face-to-face. Our team affords employees options to work in the office every day or in a flexible, hybrid work model near one of our U.S. Amazon offices.
See more jobs
View from space of a connected network around planet Earth representing the Internet of Things.
Get more from Amazon Science
Subscribe to our newsletter

Amazon.com | Conditions of Use | Privacy | © 1996-2026 Amazon.com, Inc. or its affiliates

Social