VP of Engineering, Products & Solutions

The world will never stop evolving around your business. Whether you’re facing a tough system design challenge on an engineering level or a strategic problem from a company leadership perspective, this is an inescapable truth. Our mission at Xoople is to equip companies with the trusted, verifiable Earth data they need to track such unavoidable change and be successfully adaptive.

Right now, too many organizations are relying on advanced AI models that have incredible linguistic intelligence but are ineffective at interpreting the world in a robust, repeatable, and systemic way. Users open themselves up to risk by relying on outputs that seem confident but are misrepresentative of reality.

Take the example of a railway company building a new, cross-country rail network that will span hundreds of miles across multiple regions. Nothing about that process will remain static. Extreme flooding across the corridor can send building schedules into a tailspin. Shipping lanes delivering vital supplies can become unpassable. All these factors threaten its ability to deliver the project – and the value that was promised – to its stakeholders.

How can this railway leadership team make timely, high-stakes decisions if their digital planning systems rely on outdated data, making them blind to what is physically happening on the ground each day?

At the recent Microsoft Build event in San Francisco, I sat down with Azure’s Nate Waters on BuildTV to talk about the data, context, and AI infrastructure required to bridge this gap and Xoople’s approach to the engineering challenge that lies behind Earth’s System of Record™.

Building Earth’s foundational data layer

Making physical-world intelligence easily consumable for enterprise workflows requires the processing of trillions of pixels. We work with massive, heavy multimodal images—often hundreds of megabytes each—containing 10, 15, or more spectral bands. To transform this raw, complex imagery into structured, explainable variables that an enterprise risk tool or an AI agent can consume, we’ve designed an engineering approach that can deliver the scale and speed customers need, including:

• Building with Anyscale on Azure: Incorporating Anyscale on Azure within Xoople’s tech stack allows us to simplify the scaling of our models on top of our GPUs and CPUs. As an abstraction layer, it reduces time to outcome and enables us to focus on building our product and optimizing end user experience.
• Squeezing GPU Performance: In collaboration with Anyscale, we streamlined the CPU-to-GPU handshake. Using CPUs for data preparation and GPUs for running foundation models, we pushed GPU utilization into the high 90s. This high-efficiency pipeline is what allows us to run models like NASA and IBM’s TerraMind to extract embeddings and perform few-shot labelling at scale.
• Decomposing Problems: When scaling, our team thinks in “the smallest solvable units.” By breaking our pipelines down into independent tasks that can be issued millions of times in parallel, we build clear roadmaps for development and enable the team to scale more efficiently.

From the engineer’s screen to the customer’s system

Our engineering choices focus entirely on enabling ease at the enterprise application layer. By removing the friction and complexity of raw geospatial processing, we empower businesses to bypass heavy data engineering and focus on high-stakes decisions. Rather than forcing a supply chain manager to decipher complex, raw imagery to identify port bottlenecks, or requiring an energy team to build custom pipelines to monitor ground subsidence along a pipeline corridor, we deliver high-quality insights directly into their existing systems.

Ultimately, our engineering is what makes this seamless experience possible. We plug physical-world intelligence directly into the daily workflows our customers rely on, so they can access actionable ground truth exactly where they need it.