Striveworks was launched in 2018 with a vision to build the plumbing for AI underneath the application layer. In addition to offering the best AI-native experience for consumers of models, Striveworks also gives developers the power of a full-stack AI operations (AIOps) platform embedded in their product. AI models are in constant flux, and, at the time of Striveworks’ founding, select enterprise-grade solutions were already integrating AI at scale. Just a few years later, transformer-based models would exponentiate the adoption of AI, and the demand to abstract the complexity of AI implementation without sacrificing specificity and performance would be vastly accelerated.
The Maxar Intelligence team began exploring Striveworks’ AIOps platform and very quickly saw an opportunity to integrate it with their best-of-class imagery and abstractions around tasking and collecting.
“Coming from our decades leading geospatial intelligence, we saw very clearly that an effective persistent site monitoring product had to be a complete, end-to-end solution,” said jC Clark, VP of Insights at Maxar Intelligence. “Site monitoring isn’t just tasking a sensor today; it spans everything from archival data to AI. Our product vision demanded an AIOps scaffolding that was flexible enough to keep pace with our customers’ dynamic needs while heavily automating our model life cycle process. Striveworks is an obvious partner who can help us build these end-to-end site monitoring solutions.”
Maxar turned to Striveworks to be an AI “command center” inside its site monitoring workflows and to deliver out-of-the-box AI capabilities for object and anomaly detection to its customers. Striveworks and Maxar have built an integrated capability that allows customers to now leverage third-party foundational models and fine-tune their own proprietary models with data collected through Maxar’s virtual constellation—which includes Maxar’s own very high-resolution imagery, as well as partner imagery. This capability has been fed into solutions like SentryTM, Maxar’s global-scale predictive intelligence product.
Here, we’ll walk through how dynamic model comparison, model selection, and zero-downtime model updates came together to deliver a high-quality, AI-native site monitoring application to Maxar customers, with transparent acceleration of the underlying AIOps for the Maxar development team. Dynamic model comparison and model selection ensures that users get the best model for each point of data at inference time.
Our dynamic routing ensures that the optimal model performs inference on each data point, improving production performance.
The Maxar development team made extensive use of Striveworks’ evaluation framework to quickly build model cards and compare models. This framework enables them to customize their evaluation metrics, evaluate models against different data stratifications, and productionize the models that performed the best for their desired use cases. The evaluation process was not performed in a vacuum; it was accelerated and complemented by other key components of the Striveworks platform, such as high throughput model inferencing, data science workspaces, and dataset management.
Striveworks provides visualizations of model performance and comparison across data stratifications.
For use cases that require more bespoke analysis, Striveworks’ SDK allows users to extend the platform. In the following example, an inference run against a dataset is turned into a set of metrics that a developer can further filter and analyze.
Striveworks' software development kit (SDK) extends its evaluation framework to give developers the tools to go beyond what's in the user interface (UI) and dig into their model evaluations.
The platform stores a concise summary of model performance, along with detailed evaluation metrics and other metadata that can be queried for deeper analysis. All necessary data for full traceability and reproducibility is captured.
By integrating captured production inferences and metadata, user feedback, and model monitoring, developers found it easy to continually tune and optimize model performance—without downtime or regressions in model performance. The end result here is that consumers of model inferences get a superior experience driven by better model performance.
Finally, the partnership enabled Maxar’s customers to move into site monitoring at enterprise scale. Tens of thousands of users, thousands of sites, and hundreds of models to process and consume petabytes of imagery are nontrivial. Beyond that horizontal scaling, these customers rely on Maxar for their most critical tasks. Trust, auditability, and compliance are critical needs. The Striveworks platform’s model and data lineage ensures that compliance teams, regulators, and auditors have the information they need.
Dynamic model selection, automated evaluation pipelines, and zero-downtime model updates don’t have to be months-long engineering projects. Maxar’s success demonstrates how the right AIOps platform turns complex AI operations into simple integrations, letting developers focus on application logic instead of infrastructure. This speed to market has enabled Maxar to deliver site monitoring as a service to strategic customers.
Interested in seeing how you can build AI-native applications faster with Striveworks? Contact us, join our community, or check out the SentryTM product page.
