Social
Automated reasoning

Better-performing “25519” elliptic-curve cryptography

Automated reasoning and optimizations specific to CPU microarchitectures improve both performance and assurance of correct implementation.

By Torben Hansen, John Harrison
September 10, 2024
12 min read
Share

Cryptographic algorithms are essential to online security, and at Amazon Web Services (AWS), we implement cryptographic algorithms in our open-source cryptographic library, AWS LibCrypto (AWS-LC), based on code from Google’s BoringSSL project. AWS-LC offers AWS customers implementations of cryptographic algorithms that are secure and optimized for AWS hardware.

Two cryptographic algorithms that have become increasingly popular are x25519 and Ed25519, both based on an elliptic curve known as curve25519. To improve the customer experience when using these algorithms, we recently took a deeper look at their implementations in AWS-LC. Henceforth, we use x/Ed25519 as shorthand for “x25519 and Ed25519”.

Graviton chip.png
Related content
Formal verification makes RSA faster — and faster to deploy
Optimizations for Amazon's Graviton2 chip boost efficiency, and formal verification shortens development time.

In 2023, AWS released multiple assembly-level implementations of x/Ed25519 in AWS-LC. By combining automated reasoning and state-of-the-art optimization techniques, these implementations improved performance over the existing AWS-LC implementations and also increased assurance of their correctness.

In particular, we prove functional correctness using automated reasoning and employ optimizations targeted to specific CPU microarchitectures for the instruction set architectures x86_64 and Arm64. We also do our best to execute the algorithms in constant time, to thwart side-channel attacks that infer secret information from the durations of computations.

In this post, we explore different aspects of our work, including the process for proving correctness via automated reasoning, microarchitecture (μarch) optimization techniques, the special considerations for constant-time code, and the quantification of performance gains.

Elliptic-curve cryptography

Elliptic-curve cryptography is a method for doing public-key cryptography, which uses a pair of keys, one public and one private. One of the best-known public-key cryptographic schemes is RSA, in which the public key is a very large integer, and the corresponding private key is prime factors of the integer. The RSA scheme can be used both to encrypt/decrypt data and also to sign/verify data. (Members of our team recently blogged on Amazon Science about how we used automated reasoning to make the RSA implementation on Amazon’s Graviton2 chips faster and easier to deploy.)

Elliptic curve.png
Example of an elliptic curve.

Elliptic curves offer an alternate way to mathematically relate public and private keys; sometimes, this means we can implement schemes more efficiently. While the mathematical theory of elliptic curves is both broad and deep, the elliptic curves used in cryptography are typically defined by an equation of the form y2 = x3 + ax2 + bx + c, where a, b, and c are constants. You can plot the points that satisfy the equation on a 2-D graph.

An elliptic curve has the property that a line that intersects it at two points intersects it at at most one other point. This property is used to define operations on the curve. For instance, the addition of two points on the curve can be defined not, indeed, as the third point on the curve collinear with the first two but as that third point’s reflection around the axis of symmetry.

Elliptic-curve addition.gif
Addition on an elliptic curve.

Now, if the coordinates of points on the curve are taken modulo some integer, the curve becomes a scatter of points in the plane, but a scatter that still exhibits symmetry, so the addition operation remains well defined. Curve25519 is named after a large prime integer — specifically, 2255 – 19. The set of numbers modulo the curve25519 prime, together with basic arithmetic operations such as multiplication of two numbers modulo the same prime, define the field in which our elliptic-curve operations take place.

Successive execution of elliptic-curve additions is called scalar multiplication, where the scalar is the number of additions. With the elliptic curves used in cryptography, if you know only the result of the scalar multiplication, it is intractable to recover the scalar, if the scalar is sufficiently large. The result of the scalar multiplication becomes the basis of a public key, the original scalar the basis of a private key.

The x25519 and Ed25519 cryptographic algorithms

The x/Ed25519 algorithms have distinct purposes. The x25519 algorithm is a key agreement algorithm, used to securely establish a shared secret between two peers; Ed25519 is a digital-signature algorithm, used to sign and verify data.

The x/Ed25519 algorithms have been adopted in transport layer protocols such as TLS and SSH. In 2023, NIST announced an update to its FIPS185-6 Digital Signature Standard that included the addition of Ed25519. The x25519 algorithm also plays a role in post-quantum safe cryptographic solutions, having been included as the classical algorithm in the TLS 1.3 and SSH hybrid scheme specifications for post-quantum key agreement.

Microarchitecture optimizations

When we write assembly code for a specific CPU architecture, we use its instruction set architecture (ISA). The ISA defines resources such as the available assembly instructions, their semantics, and the CPU registers accessible to the programmer. Importantly, the ISA defines the CPU in abstract terms; it doesn’t specify how the CPU should be realized in hardware.

Secure multiparty computation.gif
Related content
Amazon's Tal Rabin wins Dijkstra Prize in Distributed Computing
Prize honors Amazon senior principal scientist and Penn professor for a protocol that achieves a theoretical limit on information-theoretic secure multiparty computation.

The detailed implementation of the CPU is called the microarchitecture, and every μarch has unique characteristics. For example, while the AWS Graviton 2 CPU and AWS Graviton 3 CPU are both based on the Arm64 ISA, their μarch implementations are different. We hypothesized that if we could take advantage of the μarch differences, we could create x/Ed25519 implementations that were even faster than the existing implementations in AWS-LC. It turns out that this intuition was correct.

Let us look closer at how we took advantage of μarch differences. Different arithmetic operations can be defined on curve25519, and different combinations of those operations are used to construct the x/Ed25519 algorithms. Logically, the necessary arithmetic operations can be considered at three levels:

  1. Field operations: Operations within the field defined by the curve25519 prime 2255 – 19.
  2. Elliptic-curve group operations: Operations that apply to elements of the curve itself, such as the addition of two points, P1 and P2.
  3. Top-level operations: Operations implemented by iterative application of elliptic-curve group operations, such as scalar multiplication.
Levels of operations.png
Examples of operations at different levels. Arrows indicate dependency relationships between levels.

Each level has its own avenues for optimization. We focused our μarch-dependent optimizations on the level-one operations, while for levels two and three our implementations employ known state-of-the-art techniques and are largely the same for different μarchs. Below, we give a summary of the different μarch-dependent choices we made in our implementations of x/Ed25519.

  • For modern x86_64 μarchs, we use the instructions MULX, ADCX, and ADOX, which are variations of the standard assembly instructions MUL (multiply) and ADC (add with carry) found in the instruction set extensions commonly called BMI and ADX. These instructions are special because, when used in combination, they can maintain two carry chains in parallel, which has been observed to boost performance up to 30%. For older x86_64 μarchs that don’t support the instruction set extensions, we use more traditional single-carry chains.
  • For Arm64 μarchs, such as AWS Graviton 3 with improved integer multipliers, we use relatively straightforward schoolbook multiplication, which turns out to give good performance. AWS Graviton 2 has smaller multipliers. For this Arm64 μarch, we use subtractive forms of Karatsuba multiplication, which breaks down multiplications recursively. The reason is that, on these μarchs, 64x64-bit multiplication producing a 128-bit result has substantially lower throughput relative to other operations, making the number size at which Karatsuba optimization becomes worthwhile much smaller.

We also optimized level-one operations that are the same for all μarchs. One example concerns the use of the binary greatest-common-divisor (GCD) algorithm to compute modular inverses. We use the “divstep” form of binary GCD, which lends itself to efficient implementation, but it also complicates the second goal we had: formally proving correctness.

Secure multiparty computation.gif
Related content
Computing on private data
Both secure multiparty computation and differential privacy protect the privacy of data used in computation, but each has advantages in different contexts.

Binary GCD is an iterative algorithm with two arguments, whose initial values are the numbers whose greatest common divisor we seek. The arguments are successively reduced in a well-defined way, until the value of one of them reaches zero. With two n-bit numbers, the standard implementation of the algorithm removes at least one bit total per iteration, so 2n iterations suffice.

With divstep, however, determining the number of iterations needed to get down to the base case seems analytically difficult. The most tractable proof of the bound uses an elaborate inductive argument based on an intricate “stable hull” provably overapproximating the region in two-dimensional space containing the points corresponding to the argument values. Daniel Bernstein, one of the inventors of x25519 and Ed25519, proved the formal correctness of the bound using HOL Light, a proof assistant that one of us (John) created. (For more on HOL Light, see, again, our earlier RSA post.)

Performance results

In this section, we will highlight improvements in performance. For the sake of simplicity, we focus on only three μarchs: AWS Graviton 3, AWS Graviton 2, and Intel Ice Lake. To gather performance data, we used EC2 instances with matching CPU μarchs — c6g.4xlarge, c7g.4xlarge, and c6i.4xlarge, respectively; to measure each algorithm, we used the AWS-LC speed tool.

In the graphs below, all units are operations per second (ops/sec). The “before” columns represent the performance of the existing x/Ed25519 implementations in AWS-LC. The “after” columns represent the performance of the new implementations.

Signing new.png
For the Ed25519 signing operation, the number of operations per second, over the three μarchs, is, on average, 108% higher with the new implementations.
Verification.png
For the Ed25519 verification operation, we increased the number of operations per second, over the three μarchs, by an average of 37%.

We observed the biggest improvement for the x25519 algorithm. Note that an x25519 operation in the graph below includes the two major operations needed for an x25519 key exchange agreement: base-point multiplication and variable-point multiplication.

Ops:sec new.png
With x25519, the new implementation increases the number of operations per second, over the three μarchs, by an average of 113%.

On average, over the AWS Graviton 2, AWS Graviton 3, and Intel Ice Lake microarchitectures, we saw an 86% improvement in performance.

Proving correctness

We develop the core parts of the x/Ed25519 implementations in AWS-LC in s2n-bignum, an AWS-owned library of integer arithmetic routines designed for cryptographic applications. The s2n-bignum library is also where we prove the functional correctness of the implementations using HOL Light. HOL Light is an interactive theorem prover for higher-order logic (hence HOL), and it is designed to have a particularly simple (hence light) “correct by construction” approach to proof. This simplicity offers assurance that anything “proved” has really been proved rigorously and is not the artifact of a prover bug.

Diagram of a circuit with a multiplicative depth of two
Related content
Building machine learning models with encrypted data
New approach to homomorphic encryption speeds up the training of encrypted machine learning models sixfold.

We follow the same principle of simplicity when we write our implementations in assembly. Writing in assembly is more challenging, but it offers a distinct advantage when proving correctness: our proofs become independent of any compiler.

The diagram below shows the process we use to prove x/Ed25519 correct. The process requires two different sets of inputs: first is the algorithm implementation we’re evaluating; second is a proof script that models both the correct mathematical behavior of the algorithm and the behavior of the CPU. The proof is a sequence of functions specific to HOL Light that represent proof strategies and the order in which they should be applied. Writing the proof is not automated and requires developer ingenuity.

From the algorithm implementation and the proof script, HOL Light either determines that the implementation is correct or, if unable to do so, fails. HOL Light views the algorithm implementation as a sequence of machine code bytes. Using the supplied specification of CPU instructions and the developer-written strategies in the proof script, HOL Light reasons about the correctness of the execution.

CI integration.png
CI integration provides assurance that no changes to the algorithm implementation code can be committed to s2n-bignum’s code repository without successfully passing a formal proof of correctness.

This part of the correctness proof is automated, and we even implement it inside s2n-bignum’s continuous-integration (CI) workflow. The workflow covered in the CI is highlighted by the red dotted line in the diagram below. CI integration provides assurance that no changes to the algorithm implementation code can be committed to s2n-bignum’s code repository without successfully passing a formal proof of correctness.

The CPU instruction specification is one of the most critical ingredients in our correctness proofs. For the proofs to be true in practice, the specification must capture the real-world semantics of each instruction. To improve assurance on this point, we apply randomized testing against the instruction specifications on real hardware, “fuzzing out” inaccuracies.

Constant time

We designed our implementations and optimizations with security as priority number one. Cryptographic code must strive to be free of side channels that could allow an unauthorized user to extract private information. For example, if the execution time of cryptographic code depends on secret values, then it might be possible to infer those values from execution times. Similarly, if CPU cache behavior depends on secret values, an unauthorized user who shares the cache could infer those values.

Our implementations of x/Ed25519 are designed with constant time in mind. They perform exactly the same sequence of basic CPU instructions regardless of the input values, and they avoid any CPU instructions that might have data-dependent timing.

Using x/Ed25519 optimizations in applications

AWS uses AWS-LC extensively to power cryptographic operations in a diverse set of AWS service subsystems. You can take advantage of the x/Ed25519 optimizations presented in this blog by using AWS-LC in your application(s). Visit AWS-LC on Github to learn more about how you can integrate AWS-LC into your application.

To allow easier integration for developers, AWS has created bindings from AWS-LC to multiple programming languages. These bindings expose cryptographic functionality from AWS-LC through well-defined APIs, removing the need to reimplement cryptographic algorithms in higher-level programming languages. At present, AWS has open-sourced bindings for Java and Rust — the Amazon Corretto Cryptographic Provider (ACCP) for Java, and AWS-LC for Rust (aws-lc-rs). Furthermore, we have contributed patches allowing CPython to build against AWS-LC and use it for all cryptography in the Python standard library. Below we highlight some of the open-source projects that are already using AWS-LC to meet their cryptographic needs.

Open-source projects.png
Open-source projects using AWS-LC to meet their cryptographic needs.

We are not done yet. We continue our efforts to improve x/Ed25519 performance as well as pursuing optimizations for other cryptographic algorithms supported by s2n-bignum and AWS-LC. Follow the s2n-bignum and AWS-LC repositories for updates.

Research areas
Tags
About the Author
Torben Hansen
Torben Hansen is an applied scientist with AWS Cryptography.
John Harrison
John Harrison is a senior principal applied scientist in Amazon’s Automated Reasoning Group. He is a maintainer of s2n-bignum and the HOL Light theorem prover.

Related content

Work with us

See more jobs See more jobs
Senior Inference Engineer, AGI
US, CA, Sunnyvale
We are looking for a Senior Inference Engineer to own inference for real-time multimodal conversational AI. This is a full-stack inference role: you will work across the entire path a model takes from research to production — shaping model architecture so it is servable, building the real-time runtime that serves it within hard latency budgets, and building the offline systems that train and reinforce it. You will operate at the boundary of Science and Inference, taking frontier-scale speech and audio models and making them run within real-time latency budgets on production hardware. You will co-design architectures with scientists to make them inference-friendly from inception, own the low-latency streaming serving path, and build the training and reinforcement-learning infrastructure that closes the loop. You will have the compute, data, and runway to solve problems that few teams in the world are positioned to tackle. As a Senior Engineer, you will own a significant area of the inference stack end to end, drive its technical execution, contribute to the team's roadmap, and work closely with scientists and hardware partners to ensure our models run fast enough to feel human in real time — and at a cost that makes them viable at scale. You may go deep in one of the areas below while contributing across the others. Key job responsibilities Model Architecture & Inference Co-Design • Partner with research scientists to make model architectures servable from inception — surfacing the latency, memory, and cost implications of architecture choices before they are locked in • Implement and optimize the inference path for large-scale multimodal models — attention and KV-cache mechanisms, multimodal/autoregressive decoding, and the compute primitives on the critical path Apply efficiency techniques across the stack — quantization (per-tensor/per-channel/per- group, INT8/FP8/BF16), speculative decoding, operator fusion, and paged KV-cache — and quantify their quality/latency trade-offs • Develop and tune high-performance kernels for critical operations where off-the-shelf implementations leave performance on the table, integrating them into production serving with minimal overhead • Profile end-to-end performance with tools such as Nsight Compute/Systems and roofline analysis to identify and eliminate bottlenecks in large-scale inference workloads Real-Time & Interactive Runtime • Own the real-time serving path for streaming multimodal conversational AI, meeting sub- second, streaming latency budgets under concurrent session load • Build and tune continuous batching, scheduling, and preemption to balance throughput against per-request latency SLAs for interactive workloads • Customize production serving frameworks (e.g., vLLM, PyTorch) for real-time streaming generative models that fall outside standard LLM serving patterns — sustained low-latency output under concurrent session load • Implement multi-GPU inference (tensor parallelism, collective communication) for latency- critical paths, and drive cost toward parity with existing production baselines • Establish latency, throughput, and cost benchmarking, and publish the operational metrics that gate deployment Offline Systems: Training, RL & Evaluation Infrastructure • Build and scale the offline inference systems behind post-training — high-throughput rollout generation and reward-model serving for reinforcement learning (RL/RLHF/RLAIF) • Ensure train/serve consistency — that the inference path used in RL and evaluation faithfully matches production online behavior (e.g., parity across sampling and logit processing) • Work with the evaluation team to enable offline inference that captures the quality dimensions unique to real-time conversation — latency sensitivity, audio quality, and interaction naturalness
Applied Scientist, Last Mile Science
US, TX, Austin
Are You Ready to Redefine How the World Receives Its Packages? What if your algorithms defined the most efficient path for millions of deliveries — every single day? At Amazon, we're building the science that makes that possible, and we're looking for exceptional scientists to help lead the way. The Last Mile Routing & Planning organization develops the software, algorithms, and tools that power the "magic" of home delivery. Our planning and routing intelligence systems drive billions of daily decisions — enabling safe, efficient, and frustration-free routes for drivers across the globe. What You'll Do In this role, you'll sit at the intersection of state-of-the-art research and real-world impact. You will: - Design and build algorithms that solve large-scale, complex logistics problems - Synthesize data from diverse sources to identify high-value business opportunities - Provide research direction and data-driven insights to guide strategic decisions - Translate complex technical approaches into clear communication for scientists, engineers, and business stakeholders - Partner closely with scientists and engineers in a collaborative, high-impact environment What You'll Work On We have an exciting and growing portfolio of research areas, including: - Routing for same-day and grocery deliveries - Planning for electric and autonomous vehicles - District-level and stop-level planning - Forecasting solutions for diverse delivery programs All of this is powered by the latest methods in Operations Research (OR), Machine Learning (ML), and Generative AI — at a truly global scale. Successful candidates will lead one or more of these problem spaces. What We're Looking For - Deep expertise in Operations Research and/or Machine Learning methods - Proven experience applying these methods to large-scale, real-world business problems - Ability to translate models into production-ready code in Python or Java - Strong communication skills — you can explain complex technical concepts to diverse audiences - A bias for action and an iterative mindset when tackling ambitious research challenges Why Amazon We're passionate about your growth. Whether you want to explore emerging technologies, take on broader scope, or accelerate your career trajectory, we'll invest in helping you get there. Our business is scaling fast — and so are the opportunities for the people who build it. If you're driven by the challenge of optimizing one of the world's most complex logistics systems and excited to see your work impact millions of customers daily, we'd love to hear from you. Key job responsibilities - Invent and design novel solutions for scientifically complex problem areas, and identify opportunities for invention within existing and new business initiatives - Deliver large-scale, high-impact solutions to complex problems in support of medium-to-large business goals - Shape the design of scientifically complex software systems, personally contributing significant portions of the critical scientific novelty - Apply mathematical optimization, machine learning, and Generative AI techniques to develop solution methodologies for in-house decision support tools and software - Research, prototype, simulate, and experiment with models — and actively participate in their production-level deployment in Python or Java - Engage with the broader scientific community by publishing research articles and participating in leading research conferences
Principal Applied Scientist, Real-Time Conversational AI , AGI
US, WA, Seattle
We are looking for a Principal Applied Scientist to drive the research and development of real-time multimodal conversational AI. You will operate across two focus areas: advancing foundation models for speech and audio, and building the post-training systems (reward modeling, reinforcement learning) that shape natural, human-like conversational behavior. You will be the expert in your area while contributing across the full model lifecycle — from pre-training and architecture design through post-training alignment and real-time deployment. You will work at the frontier of what's possible in conversational AI, with the compute, data, and runway to pursue problems that few teams in the world have the resources to tackle. As a Principal Scientist, you will set the technical direction for your research area, influence the broader roadmap, and work closely with inference engineers to ensure your models are designed for real-time production deployment from inception. Key job responsibilities Foundation Model Scaling - Build and train large-scale multimodal foundation models for real-time speech and audio generation, from architecture design through production-scale training - Advance the scaling and efficiency of conversational modes, including the relationship between data, model size, and real time performance. - Design model architectures informed by hardware constraints and inference requirements, working with inference engineers to ensure models are servable from inception - Develop training methodologies for multimodal models that jointly process and generate speech, language, and audio in real-time streaming contexts Post-Training & Reinforcement Learning - Design and build reward models and reward functions for speech systems — capturing naturalness, fluency, conversational quality, and real-time responsiveness - Develop and apply reinforcement learning methods to shape conversational behavior — teaching models natural timing, responsiveness, and fluid interaction - Build the post-training pipeline from SFT through RL alignment, optimized for real-time multimodal outputs rather than text-only generation - Design evaluation frameworks that capture the quality dimensions unique to real-time conversation Real-Time Perception & Generation - Advance the team's capabilities in real-time perception - Work at the intersection of model architecture and production constraints to ensure multimodal capabilities function within hard real-time latency budgets
Flight Dynamics Engineer, Amazon Leo Flight Dynamics
US, WA, Redmond
We are searching for a talented candidate with experience in orbital mechanics, orbit determination, launch vehicle trajectories, and launch vehicle mission planning. In this position, a successful candidate would serve as a Research Scientist in support of Amazon Leo’s constellation with particular focus on Launch Vehicle support. Strong analysis skills are required to develop engineering studies of complex large-scale dynamical systems. This position requires demonstrated expertise in computational analysis automation and tool development. Export Control Requirement: Due to applicable export control laws and regulations, candidates must be a U.S. citizen or national, U.S. permanent resident (i.e., current Green Card holder), or lawfully admitted into the U.S. as a refugee or granted asylum. Key job responsibilities Working with the Leo GNC team, you will: • Perform spacecraft maneuver or navigation analysis in support of multi-disciplinary trades within the Amazon Leo team. • Contribute to prototype software development of flight algorithms. • Test and assess navigation software for integration into flight systems. • Assess and trouble-shoot the performance of Leo on-board GNSS hardware and software systems. • Work closely with GNC engineers to manage on-orbit performance and develop flight dynamics operations processes. • Manage engineering trades as needed for various launch vehicle mission designs. • Support Leo’s Launch Vehicle Mission Management team with technical expertise in Launch Vehicle trajectory requirements specification • Evaluate Launch Vehicle performance and compliance with mission requirements • Develop tools to support Mission Management planning for over 80 launches! • Work collaboratively with launch vehicle system technical teams About the team The Flight Dynamics team is responsible for the guidance, navigation, control and safety of the spaceflight of the Amazon Leo constellation. This team provides solutions to spaceflight challenges in constellation design, orbit selection, launch vehicle insertion requirements, navigation, trajectory design, space situational awareness, and space traffic coordination.
Applied Scientist II, Amazon Robotics
DE, Berlin
The Amazon Robotics team is seeking an experienced Applied Scientist to join our team. In this role you will apply the latest trends in research to solve real-world problems in robotics and AI. You will collaborate with a team of scientists and engineers building these applications. We holistically design, build, and deliver end-to-end robotic systems. Our team is also responsible for core infrastructure and tools that serve as the backbone of our robotic applications, enabling roboticists, machine learning scientists, software engineers, and hardware engineers to collaborate and deploy systems in the field. Key job responsibilities • Research, design, develop, and evaluate complex perception, motion planning, and decision making algorithms integrating across multiple disciplines and leveraging machine learning. • Create experiments and prototype implementations of new learning algorithms and prediction techniques. • Work closely with software engineering team members to drive scalable, real-time implementations. • Collaborate with machine learning and robotic controls experts to implement and deploy algorithms, such as machine learning models. • Collaborate closely with hardware engineering team members on developing systems from prototyping to production level. • Represent Amazon in academia community through publications and scientific presentations. • Work with stakeholders across hardware, science, and operations teams to iterate on systems design and implementation. About the team Watch this video to learn more about Vulcan Pick team in Amazon Robotics: https://www.amazon.science/publications/vulcan-pick-a-robotic-system-for-picking-targeted-objects-from-fabric-pods
Data Scientist, Alexa AI
IN, KA, Bengaluru
Amazon is looking for a passionate, talented, and inventive Data Scientist with machine learning background to help build industry-leading Speech and Language technology. Our mission is to provide a delightful experience to Amazon’s customers by pushing the envelope in Automatic Speech Recognition (ASR), Natural Language Understanding (NLU), Machine Learning (ML). Key job responsibilities Key job responsibilities Amazon is looking for a passionate, talented, and inventive Data Scientist with machine learning background to help build industry-leading Speech and Language technology. Our mission is to provide a delightful experience to Amazon’s customers by pushing the envelope in Automatic Speech Recognition (ASR), Natural Language Understanding (NLU), Machine Learning (ML) and Computer Vision (CV). As part of our AI team in Amazon AWS, you will work alongside internationally recognized experts to develop data experiments, novel algorithms and data techniques to advance the state-of-the-art in human language technology. Your work will directly impact millions of our customers in the form of products and services that make use of speech and language technology. You will gain hands on experience with Amazon’s heterogeneous speech, text, and structured data sources, and large-scale computing resources to accelerate advances in spoken language understanding.
Applied Scientist , Prime Video Personalization and Discovery
US, WA, Seattle
Prime Video is a first-stop entertainment destination offering customers a vast collection of premium programming in one app available across thousands of devices. Prime members can customize their viewing experience and find their favorite movies, series, documentaries, and live sports – including Amazon MGM Studios-produced series and movies; licensed fan favorites; and programming from Prime Video add-on subscriptions such as Apple TV+, Max, Crunchyroll and MGM+. All customers, regardless of whether they have a Prime membership or not, can rent or buy titles via the Prime Video Store, and can enjoy even more content for free with ads. Are you interested in shaping the future of entertainment? Prime Video's technology teams are creating best-in-class digital video experience. As a Prime Video technologist, you’ll have end-to-end ownership of the product, user experience, design, and technology required to deliver state-of-the-art experiences for our customers. You’ll get to work on projects that are fast-paced, challenging, and varied. You’ll also be able to experiment with new possibilities, take risks, and collaborate with remarkable people. We’ll look for you to bring your diverse perspectives, ideas, and skill-sets to make Prime Video even better for our customers. With global opportunities for talented technologists, you can decide where a career Prime Video Tech takes you! We are looking for a self-motivated, passionate and resourceful Applied Scientist to bring diverse perspectives, ideas, and skill-sets to make Prime Video even better for our customers. You will spend your time as a hands-on machine learning practitioner and a research leader. You will play a key role on the team, building and guiding machine learning models from the ground up. At the end of the day, you will have the reward of seeing your contributions benefit millions of Amazon.com customers worldwide. Key job responsibilities Develop AI solutions for various Prime Video Personalization systems using Deep learning, GenAI, Reinforcement Learning, and optimization methods; Work closely with engineers and product managers to design, implement and launch AI solutions end-to-end; Design and conduct offline and online (A/B) experiments to evaluate proposed solutions based on in-depth data analyses; Effectively communicate technical and non-technical ideas with teammates and stakeholders; Stay up-to-date with advancements and the latest modeling techniques in the field; Publish your research findings in top conferences and journals. About the team Prime Video Personalization and Discovery team owns science solution to power personalized experience on various devices, from sourcing, relevance, ranking, to name a few. We work closely with the engineering and product teams to launch our solutions in production.
PhD Econ Intern, Store Economics and Science
US, WA, Seattle
We are looking for detail-oriented, organized, and responsible individuals who are eager to learn how to apply their causal inference / structural econometrics skillsets to solve real world problems. The intern will work in the area of Store Economics and Science (SEAS) and develop models to SEAS. Our PhD Economist Internship Program offers hands-on experience in applied economics, supported by mentorship, structured feedback, and professional development. Interns work on real business and research problems, building skills that prepare them for full-time economist roles at Amazon and beyond. You will learn how to build data sets and perform applied econometric analysis collaborating with economists, scientists, and product managers. These skills will translate well into writing applied chapters in your dissertation and provide you with work experience that may help you with placement. About the team The Stores Economics and Science Team (SEAS) is a Stores-wide interdisciplinary team at Amazon with a "peak jumping" mission focused on disruptive innovation. The team applies science, economics, and engineering expertise to tackle the business's most critical problems, working to move from local to global optima across Amazon Stores operations. SEAS builds partnerships with organizations throughout Amazon Stores to pursue this mission, exploring frontier science while learning from the experience and perspective of others. Their approach involves testing solutions first at a small scale, then aligning more broadly to build scalable solutions that can be implemented across the organization. The team works backwards from customers using their unique scientific expertise to add value, takes on long-run and high-risk projects that business teams typically wouldn't pursue, helps teams with kickstart problems by building practical prototypes, raises the scientific bar at Amazon, and builds and shares software that makes Amazon more productive.
Applied Scientist II, International Machine Learning
IN, KA, Bengaluru
Do you want to join an innovative team of scientists who use machine learning and statistical techniques to create state-of-the-art solutions for providing better value to Amazon’s customers? Do you want to build and deploy advanced ML systems that help optimize millions of transactions every day? Are you excited by the prospect of analyzing and modeling terabytes of data to solve real-world problems? Do you like to own end-to-end business problems/metrics and directly impact the profitability of the company? Do you like to innovate and simplify? If yes, then you may be a great fit to join the Machine Learning team for India Consumer Businesses. Machine Learning, Big Data and related quantitative sciences have been strategic to Amazon from the early years. Amazon has been a pioneer in areas such as recommendation engines, ecommerce fraud detection and large-scale optimization of fulfillment center operations. As Amazon has rapidly grown and diversified, the opportunity for applying machine learning has exploded. We have a very broad collection of practical problems where machine learning systems can dramatically improve the customer experience, reduce cost, and drive speed and automation. These include product bundle recommendations for millions of products, safeguarding financial transactions across by building the risk models, improving catalog quality via extracting product attribute values from structured/unstructured data for millions of products, enhancing address quality by powering customer suggestions We are developing state-of-the-art machine learning solutions to accelerate the Amazon India growth story. Amazon India is an exciting place to be at for a machine learning practitioner. We have the eagerness of a fresh startup to absorb machine learning solutions, and the scale of a mature firm to help support their development at the same time. As part of the India Machine Learning team, you will get to work alongside brilliant minds motivated to solve real-world machine learning problems that make a difference to millions of our customers. We encourage thought leadership and blue ocean thinking in ML. Key job responsibilities Use machine learning and analytical techniques to create scalable solutions for business problems Analyze and extract relevant information from large amounts of Amazon’s historical business data to help automate and optimize key processes Design, develop, evaluate and deploy, innovative and highly scalable ML models Work closely with software engineering teams to drive real-time model implementations Work closely with business partners to identify problems and propose machine learning solutions Establish scalable, efficient, automated processes for large scale data analyses, model development, model validation and model maintenance Work proactively with engineering teams and product managers to evangelize new algorithms and drive the implementation of large-scale complex ML models in production Leading projects and mentoring other scientists, engineers in the use of ML techniques About the team International Machine Learning Team is responsible for building novel ML solutions that attack India first (and other Emerging Markets across MENA and LatAm) problems and impact the bottom-line and top-line of India business. Learn more about our team from https://www.amazon.science/working-at-amazon/how-rajeev-rastogis-machine-learning-team-in-india-develops-innovations-for-customers-worldwide
Applied Scientist II, Amazon Recommerce India
IN, KA, Bengaluru
Every product a customer returns is a moment where Amazon either recovers value or writes it off — and India's ReCommerce business is on a multi-million-dollar mission to recover more of it, more intelligently, at scale. Machine learning is the core lever: predicting whether a returned unit is sellable without a human touching it, detecting damage and fraud inside sealed packaging from images, routing each unit to its highest-value disposition, and pricing recovered inventory dynamically. India's returns network is large, fast-growing, and structurally different from other geographies — a rich, high-impact environment for an Applied Scientist to build models that move real financial and customer-experience metrics. We are hiring an Applied Scientist to build and adapt the ML that powers India ReCommerce. You will work at the intersection of two mandates: building India-first models for problems unique to our market, and adapting proven Worldwide models to India's data, catalog, and operational reality — recalibrating them where distribution, language, and process differ. You will own problems end-to-end, from framing and data through modeling, evaluation, and production deployment, partnering closely with engineering, product, and operations. Key job responsibilities Build ML models for automated returns grading — predicting the salability of returned units from structured and unstructured signals so units can be evaluated with zero or minimal human touch, improving speed, accuracy, and recovery value. Develop computer-vision models for defect detection, condition assessment, and anomaly/fraud identification (including inside sealed packaging), and for establishing chain-of-custody and damage attribution across the returns journey. Build disposition-prediction and routing models that direct each unit to its highest-value recovery path (resale, repair, liquidation, donation, recycle) as early as possible in the network. Develop pricing and recovery-optimization models for liquidation and resale, moving from flat rates toward dynamic, grade- and condition-aware pricing. Adapt Worldwide ML models to India — retraining, recalibrating, and re-evaluating for India's return distribution, catalog, languages, and operational constraints, and closing the gaps that prevent a direct lift-and-shift. Own the full model lifecycle — problem framing, data pipelines, feature engineering, training, offline/online evaluation, monitoring, and retraining — with rigorous attention to calibration, drift, and business-metric impact. Partner cross-functionally with engineering (to productionize), product (to frame problems and measure impact), and operations (to ground models in how the network actually runs), and use modern GenAI/LLM tooling to accelerate research and delivery. A day in the life You start by reviewing the performance of a grading model in production — checking calibration and drift against last week's returns, and confirming the recovery-value lift is holding. Mid-morning, you dig into a computer-vision problem: improving detection of a damage type that's driving write-offs, using images captured across the returns journey. In the afternoon you work with a Worldwide science team to bring one of their models to India — scoping what retraining and recalibration India's data requires — then pair with an engineer to move your latest model toward production behind a clean evaluation gate. You close by framing a new problem with a product partner: quantifying the opportunity, defining the label and success metric, and sketching the modeling approach. About the team India ReCommerce owns the systems and science that turn returned and unsellable inventory into recovered value and a better customer experience. You will join a team building an increasingly automated, ML-driven returns network — leveraging Worldwide platforms where they fit and building India-first capabilities where they don't. It is a high-ownership environment with a direct line from your models to measurable business and customer outcomes.
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