Encord migrated to Google Cloud and built a platform that continuously processes data and identifies the critical events necessary for refining AI models. For customers building physical AI—such as autonomous delivery drones—this means managing huge, multimodal data streams. Autonomous vehicles (AVs) are not programmed, they are trained. Their cameras and sensors capture millions of data points, but without data tagging, labeling, and annotation, they’re unable to understand the data they ingest. Annotation enables AVs to understand the data they collect, bridging the gap between raw sensor data and human-like understanding. As such, it’s the backbone of AV safety and performance.

Part of this process requires developers to identify edge cases where the AI made a mistake. Manually finding all these critical moments in petabytes of raw video footage is practically impossible. Encord eliminates this burden. Its platform runs sophisticated internal AI models to automatically select the relevant footage and label the errors within it, saving customers considerable time and effort. With this curation process, Encord helps customers reduce their total dataset size by 35% by removing uninformative data, saving significantly on labeling and compute costs.

This automation depends on the immense computing power of Cloud GPUs running on Compute Engine. With a variety of readily available GPU chipsets to choose from, spanning cost-effective options for inference to high-performance accelerators for demanding training workloads, Encord can prototype rapidly, while continuously optimizing for cost and performance efficiency. The regional availability of these processors, meanwhile, allows Encord to deliver its services at speed, wherever its customers are. For companies operating robots or vehicles, moving data halfway across the world to be processed isn’t an option, as it results in latency and high data transfer costs. With its global footprint, Google Cloud allows Encord to process data right where it resides, keeping operations fast, while remaining compliant with local data regulations.

“Customers don't want to have to wait for data to be transferred across the Atlantic—they want it processed as fast as possible,” explains Landau. “Google Cloud allows us to bring the compute power to the data, not the other way around.”

Beyond robotics, Encord is also transforming healthcare data. The training challenge for AVs is the same for medical imaging. An AI model looking at an X-ray image needs to be trained to understand what it’s looking at. Data annotation is the process of teaching the model to be able to understand and interpret medical images.

Since migrating to Google Cloud, Encord has grown rapidly, expanding from a two-person founding team to a workforce of over 100 people. In the last year alone, the company’s usage of Google Cloud has grown by 350%, enabling Encord to rapidly scale up to help customers index and curate petabytes of multimodal data.

By building reliable data pipelines on this infrastructure, Encord’s customers can iterate faster, achieving 60% faster model training and evaluation. Furthermore, the use of high-quality, curated datasets has led to an improvement of over 20% in model performance.

As more companies recognize the importance of high-quality data for their AI solutions, Encord is seeing significant expansion across a wide range of applications, particularly in robotics and autonomous systems. To continue to meet that growing thirst for high-quality data, Encord plans to further expand its use of Google Cloud, including experimenting with Gemini to speed up its internal coding and development.

With its scalable, flexible foundation, Google Cloud has enabled Encord to stop worrying about infrastructure and focus on its mission: empowering the builders of the next generation of AI. “AI is enjoying explosive growth and we need to continue scaling up to meet that demand,” adds Landau. “Google Cloud has supported us from small startup to global scaleup and we intend to keep leaning on Google Cloud to continue that scaling journey.”