LiveEO streamlines its MLOps and achieves a 65% cost reduction for its geospatial workloads

TL;DR

When making critical decisions, many businesses forget to look up. Earth observation (EO) data obtained from satellite imagery is a vast and underused resource that can hold tremendous insights for businesses. LiveEO uses AI to draw value from that data for its infrastructure customers. 

By integrating Anyscale and Prefect into its data science operations, LiveEO achieved an optimization of up to 65% in its geospatial AI workloads, reducing both costs and runtime by significant margins and streamlining its complex data pipelines. With these improvements, LiveEO is providing better service to its customers, who maintain critical infrastructure such as electricity and transportation. For these customers, the change equates to more reliable service, quicker product updates, and safer operations.

Overview

LiveEO unlocks the full potential of satellite data to protect the public and the planet.

Organizations that own and maintain expensive, critical infrastructure such as electric grids, railway networks, and pipeline operations need the power of machine learning (ML) to remotely manage their assets for resilience, performance, and public safety. 

LiveEO specializes in building such solutions. The German startup processes large-scale satellite imagery and uses AI to turn EO data into actionable insights so that its customers can better manage their infrastructure. On any given day, LiveEO processes up to two petabytes of data in AI workloads.

With LiveEO’s solutions, operators can monitor changing conditions, stay ahead of initiatives like vegetation management, and respond quickly to events like ground movement and rapidly approaching storms. And by responding faster, operators are better able to protect their local communities by minimizing the risk of service disruptions as well as safety and environmental failures.

The Challenge

Disparate tools and tribal knowledge prevent scalability.

The LiveEO team could not efficiently or quickly scale as needed to accommodate growing EO data volumes and customer demand. The core obstacle was the company's bespoke “artisanal” data science stack that consisted of diverse tools, including Jupyter Notebook, Airflow, and AWS Batch. This, in turn, prevented standardization and made reproducibility and scalability difficult. 

LiveEO also struggled with pipeline observability, which made failure recovery cumbersome and slowed down time-to-market for new solutions. The system also presented a longer-term risk for the company because it relied on tribal knowledge that wasn’t open and accessible as team members changed — which compounded the scalability problem.

The LiveEO team set several goals to address these challenges:

• Build standardized, automated, versioned, and reproducible data science pipelines.

• Require release packages to be consumed via API or a UI and make them composable.

• Abstract the developer experience away from the infrastructure and allow for workloads to scale for CPUs and GPUs seamlessly, all in Python.

• Automate, test, and deploy packages and pipeline updates continuously.

The Solution

Anyscale sets a foundation for standardized, scalable, AI-powered data science practices.

Given the amount of data LiveEO processes in its AI workloads, the team needed to build a Python-based solution. LiveEO leveraged a combination of Anyscale’s distributed computing capabilities and Prefect’s intuitive workflow automation and orchestration engine. Anyscale provided the robust AI infrastructure to manage and scale Ray workloads, while Prefect contributed automation and monitoring capabilities. 

LiveEO gained the ability to standardize, automate, and reproduce its data science practices. Workloads can now scale seamlessly for CPU and GPU needs, all within a simple Python-based interface.

The Impact

Post-integration, LiveEO experienced dramatic improvements to the time, effort, and resources required to run and scale its workloads:

• Operational efficiency and team productivity: Workloads' runtime for LiveEO's data science team dropped by 30%, which lowered costs by 65% and increased productivity.

• Standardization and automation: The team no longer bears the time-consuming burden of managing complex, hand-crafted pipelines.

• Enhanced scalability: With the new system in place, LiveEO efficiently scaled its operations and can even go from 0 to 1,000 cores in about a minute.

• Faster, more reliable customer service: The standardization and increased efficiency allowed LiveEO to deliver faster and more reliable services to its customers so they, in turn, can better protect their investments, their reputations, the public, and the natural environment.

The Anyscale and Prefect integration has been a game-changer for LiveEO. With AI driving a far more efficient and scalable data science operation, LiveEO and its infrastructure customers are putting earth observation data to work to build safer, more reliable, and more sustainable critical infrastructure.