Amazon SageMaker Ground Truth customers

There’s a lot of labeling and annotating that is needed to manage our LLM outputs and receive high-quality data within our guaranteed SLAs. Amazon SageMaker Ground Truth Plus has become an intrinsic part of our LLMs.

It's always important to have human validation, or a human in the loop, that helps you steer the models toward the right direction.

With Amazon SageMaker Ground Truth we were able to improve our business operations by generating and summarizing data from documents and speed up time to market, and save hours of productivity, and achieve higher accuracy and performance with responses. By utilizing the Ground Truth Plus workforce we were able to launch a GenAI powered question and answering and summarization for our employees and customers - that was aligned with human preferences.

Getting high quality labeled datasets is essential for the success of our generative AI models. Thanks to Amazon SageMaker Ground Truth Plus, we were able to quickly generate labeled videos and accelerate the process of fine tuning our generative AI models. Until today, our attempts at building an in house data labeling UI were expensive and time consuming, and took our data scientists over an hour a day to label the data. With Amazon SageMaker Ground Truth Plus we were able to scale from 100 to 100K of high quality labeled videos in 1 month instead of 1 year. This saved our data scientists an estimated 1000 hours, $200K in costs, and significantly enhanced our team's productivity, as well as the quality and accuracy of our generative AI models.

Our machine learning–based multi-modal perception algorithms need to be continuously trained and validated using large amounts of labeled data to ensure a reliable and safe driving system. We use Amazon SageMaker Ground Truth Plus data labeling services to label thousands of miles of real-world data, which allows us to train these models with extremely high confidence. Thanks to Amazon SageMaker Ground Truth Plus, we were able to sustain our aggressive development timelines and label millions of real-world objects, which were delivered with a quality target of 99% or greater. These time and quality-sensitive deliveries yielded an 8% improvement in precision and a 2% improvement in recall for our multi-sensor learned perception model.

At the NFL, we continue to look for new ways to use machine learning to help our fans, broadcasters, coaches, and teams benefit from deeper insights. Football is a fast moving sport where plays can happen in a split second. While coaches and referees carefully watch the game, it can be difficult to watch all players on a field for safety. Computer vision allows us to accurately detect player safety incidents, but developing these algorithms requires expertly labeled data. Now with Amazon SageMaker Ground Truth Plus, we have custom workflows and user interfaces for sophisticated labeling tasks, which helps us improve player safety.

At Airbnb, we are increasingly integrating ML across all aspects of our business. As a result, our teams consistently need to generate and maintain high-quality data in order to train and test ML models. We were looking for a way to generate high quality text classification data results on one hundred thousand paragraphs of customer service logs in Mandarin so we can better serve our customers and reduce dependencies on our customer service team. With Amazon SageMaker Ground Truth Plus, the AWS Team built a customized data labeling workflow, which included a customized ML model that was able to achieve 99% classification accuracy.

At Samsara, we’re driving the digital transformation of physical operations. With our Connected Operations Cloud, companies that depend on physical operations can harness IoT data, analytics, and AI to develop actionable insights and improve their safety, efficiency, and sustainability. With Amazon SageMaker Ground Truth Plus, we’re able to receive high quality labeled data and gain access to purpose-built tooling designed to further improve our ability to detect and address safety risks.

At VIZIO, we consistently look for ways to leverage ML to create personalized experiences for our customers. We were looking for a way to continuously review ad videos and generate commercial metadata for efficient ads classification. With the use of Amazon SageMaker Ground Truth Plus’s streaming capability, we can now use a custom template which provides video classification, metadata collection, and an automated system that enables data collection in real time as ads air. With Amazon SageMaker Ground Truth Plus we are able to review the results in less than 1 business day.

For us, machine learning brings light to unseen challenges. In the US alone, each year billions of dollars are spent cleaning up litter. With computer vision models, we transform images of litter all around the world into data, so cities can better allocate their litter management resources. However, building object detection models requires access to object, material, and brand information, as well as localized knowledge due to datasets being spread across the globe. Amazon SageMaker Ground Truth Plus allows us to create a hierarchical annotation interface that capture these precise features within that localized context. In addition, the SageMaker Ground Truth Plus expert workforce created localized image annotations, which provides a standardized solution increasing our data labeling efficiency by up to 20%, accelerating our ability to ingest annotated results into our database by 200%, and reducing post-processing time by 90%.

Our goal is to enable Canvas AMRs to navigate in a dynamic warehouse environment, keep track of surrounding obstacles and moving objects, and plan a safe and efficient route to their destination. To achieve this, it is critical to acquire 3D tracking annotations of moving objects around the robot on a large scale efficiently. We worked with Amazon ML Solutions Lab to build a scalable 3D point cloud object tracking pipeline using Amazon SageMaker Ground Truth in just a few weeks. It was impressive to see this pipeline can reduce labeling time by 6-10x and accelerate our annotation progress.

PrecisionHawk is a leading provider of drone technology for the enterprise. Our end-to-end platform leverages A.I. and machine learning to turn aerial data into actionable business intelligence. As part of this solution, we are training custom models to identify critical objects and anomalies to improve the precision and speed of critical asset inspections. To generate the training datasets for these models, we need to label an extensive corpus of data and ensure the labels are accurate. Amazon SageMaker Ground Truth is instrumental in helping us achieve our objectives here. First, it provides an intuitive user interface to kick off labeling jobs to get started quickly. In addition, the service offers the extensibility to design and deploy customer specific labeling workflows. Amazon SageMaker Ground Truth will continue to be an important part of our AI initiatives going forward.

AstraZeneca has been experimenting with machine learning across all stages of research and development, and most recently in pathology to speed up the review of tissue samples. The machine learning models first learn from a large, representative data set. Labeling the data is another time-consuming step, especially in this case, where it can take many thousands of tissue sample images to train an accurate model. AstraZeneca uses Amazon SageMaker Ground Truth, a machine learning-powered, human-in-the-loop data labeling and annotation service to automate some of the most tedious portions of this work, resulting in reduction of time spent cataloging samples by at least 50%.

The AI @ T-Mobile team is integrating AI and machine learning into the systems in our customer care centers enabling our Team of Experts to serve customers with greater speed and accuracy through Natural Language Understanding models that show them relevant, contextual customer information in real-time. Labeling data has been foundational to creating high performing models, but is also a monotonous task for our data scientists and software engineers. SageMaker Ground Truth makes the data labeling process easy, efficient and accessible, freeing up time for them to focus on what they love – building products that deliver the best experiences for our customers and care representatives.

Pinterest is continuously developing machine learning systems to detect objects for visual search and moderation use cases. To accomplish this, we need to label millions of images to generate the required training datasets. Pinterest has an existing labeling platform that has integrated Amazon services like Amazon Mechanical Turk. We were excited to explore using SageMaker Ground Truth to extend this platform to support bounding box labeling tasks. We found SageMaker Ground Truth provides a simple, streamlined interface to kick off labeling jobs. We worked closely with the AWS team to tailor SageMaker Ground Truth to our unique dataset, and look forward to integrating SageMaker Ground Truth with our data labeling platform.

Change Healthcare, a leading healthcare technology company, plays a vital role in helping the healthcare ecosystem not only function, but also work smarter. Our AI team is looking for a solution that can label text paragraphs efficiently, so we can annotate highly unstructured health data that previously could not be modeled. With SageMaker Ground Truth and its integration with SageMaker, it is easy to use with rapid workforce deployments and enables us to launch labeling tasks with very little effort, ultimately helping us to make the healthcare system more efficient.

AWS continues to show a tremendous commitment to enabling machine learning for all developers. SageMaker Ground Truth consolidates the fragmented landscape of data labeling services with a simple and well executed labeling solution. We were quickly able to integrate the tool into our training pipeline and are excited to see how its evolution will further impact our business.

We specialize in building AI solutions and bringing them to our customers to solve their business problems. We believe SageMaker Ground Truth will become a key part of our efforts towards delivering cutting-edge AI solutions for our customers. It provides a number of powerful capabilities that help us generate accurate training datasets. The “bring your own labeling workforce” with the choice of labeling templates enables us to securely onboard our team with ease and at scale. We are looking forward to using SageMaker Ground Truth across our AI solution portfolio.

The rise of AI has transformed how employers source talent and job seekers find work. ZipRecruiter’s AI-powered algorithm learns what each employer is looking for and provides a personalized, curated set of highly relevant candidates. On the other side of the marketplace, the company’s technology matches job seekers with the most pertinent jobs. And to do all that efficiently, we needed a Machine Learning model to extract relevant data automatically from uploaded resumes. Training a Machine Learning model to be able to identify the most important information requires a sizable dataset to start. The process to create this data is often expensive, manual, and time-consuming. Amazon SageMaker Ground Truth will significantly help us reduce the time and effort required to create datasets for training. Due to the confidential nature of the data, we initially considered using one of our teams but it would take time away from their regular tasks and it would take months to collect the data we needed. Using Amazon SageMaker Ground Truth, we engaged iMerit, a professional labeling company that has been pre-screened by Amazon, to assist with the custom annotation project. With their assistance we were able to collect thousands of annotations in a fraction of the time it would have taken using our own team.

At Tyson Foods, we are engaged in the production of food, seeking to pursue truth and integrity, and committed to creating value for our shareholders, our customers, our team members, and our communities. To deliver on this promise and provide the highest quality products to customers, we have manual inspections in place to detect product quality issues, including breading voids, burns, or deformations, and equipment inspections, such as conveyor belt leaks, to detect issues early; however, because these issues are anomalous, it's very difficult or impossible to collect images for machine learning model training. In addition, we often run into a bottleneck of labeling training data specific to our processes and environment. In some cases, labeling is very tedious and error prone, leading to poor machine learning model performance. Amazon SageMaker Ground Truth holds tremendous promise for us because it allows us to address every one of these challenges. Generating synthetic data will allow us to train highly accurate models to automate product and equipment inspection points. It can also reduce the turnaround time for labeled data, enabling us to train models faster while also improving accuracy. SageMaker Ground Truth is opening up paths to tackle use cases that were all but impossible to address with computer vision in the past due to the lack of example data.

Our customers need to achieve more picks per day. Facing labor shortage and volume surge, they often want to onboard induction and other solutions quickly to meet demands. Plus One Robotics’ technologies are built to tackle the variability of items handled in the warehouse. New SKUs may only show up rarely at first, and in certain circumstances. Using synthetic data allows us to pretrain our systems to deal with the wide variety of anticipated scenarios in which we will eventually encounter them. We use Amazon SageMaker Ground Truth to generate tens of thousands of labeled, photorealistic images of packages from various carriers, modeling pose, placement, and even surface variations like stickers or labels. These images have demonstrated improvements in our AI vision. Synthetic images allow us to train better performing systems faster, by virtue of their completeness and exactness, enabling us to skip the tedious data annotation and cleaning steps. This data is used in real life induction applications to facilitate over 1 million picks per day for our customers.

The University of Washington and Amazon launched a collaboration known as the Science Hub to advance artificial intelligence and robotics in open research. For one of our projects, we are using machine learning (ML) to build 3D models of the world based on a single camera viewpoint, which would enable robotic systems to understand their environment. However, getting access to large volumes of real-world training data is always a challenge for ML projects. We are excited about the promise of Amazon SageMaker Ground Truth, so we can quickly and easily generate thousands of labeled, photorealistic images, without the time-consuming and error-prone process of manually labeling images. In the early stages of our research, SageMaker Ground Truth will enable our research staff to focus on the core competency and main research objective of developing new technology, rather than dataset curation.