A quarterly snapshot of our experimental tools and open-source innovations, showcasing the projects that drive our curiosity forward.
Innovation is at the core of what we do. We love tackling complex problems with fresh ideas, whether it’s rethinking how STAC APIs work, making AI more useful in our workflows, or pushing the limits of GPU-powered geospatial analysis. Sometimes, these ideas turn into full-fledged tools, and other times, they serve as stepping stones for the broader open-source community.
1. A stac-fastapi server implementation with a stac-geoparquet backend
We’re exploring an alternative approach to STAC APIs by leveraging stac-geoparquet, providing a lightweight, object-storage-based backend that doesn’t require a traditional database. This could be especially useful for organizations that need to efficiently search and analyze geospatial imagery without the burden of maintaining complex infrastructure. Our proof-of-concept demonstrates how stac-geoparquet can power a stable, cost-effective, and scalable API, making geospatial data more accessible to those without dedicated software engineering teams. We’ll be sharing insights and inviting contributions as we refine this approach.
Check out the repo here.
Pete approved.
2. AI-aided Knowledge Management
We’ve been experimenting with building an agentic internal knowledge management tool. We called it Moscatel after the sweet Portuguese wine we were drinking when we came up with the idea. Building Moscatel gave us valuable insights into the challenges and opportunities of integrating LLM-powered knowledge management into our workflows. While we successfully created a system that could pull contextual information from GitHub issues, project documentation, and other internal sources, we encountered key limitations such as performance bottlenecks, security concerns, and hallucinations. Considering the rapid evolution of commercial AI assistants in the tools we already use, we decided that, rather than duplicating efforts, our energy would be better spent on developing modular, composable AI components that can enhance specific workflows across internal and external projects.
This shift in approach means focusing on AI-driven geospatial tools, such as natural language interfaces for STAC queries, AI-assisted storytelling for VEDA, and automated metadata generation. These components will be reusable across different platforms, making AI more accessible and adaptable to diverse use cases. While we’re pausing further development on Moscatel as a standalone tool, we rolled it out for internal beta testing and are exploring third-party integrations that align with our needs. This experience reinforced our understanding of building AI solutions that are flexible, practical, and directly embedded into real-world workflows.
Knowlege management ages like a fine wine.
3. Los Angeles Wildfire Map & Data Resources
In response to the increasing need for accessible wildfire data, we were motivated to quickly compile a Los Angeles Wildfire Map and Data Resources spreadsheet to support emergency responders, researchers, and decision-makers. This easy-to-use resource curates dashboards, satellite imagery, and datasets to aid people in understanding, monitoring, and recovering from wildfires in Southern California. It provides details on access requirements, licensing, and applicability so users can better understand this critical information. We originally shared this with Mutual Aid LA, who is coordinating massive volunteer and community-led response efforts. We’ll continue updating it as new resources become available. Check it out and let us know if there’s more we should add!
Sometimes a simple solution is the best.
4. STAC Auth Proxy
One of the biggest challenges in operationalizing STAC is authentication and authorization (auth*n), which ensures users can access only the data they are allowed to see while maintaining flexibility across different deployments. STAC Auth Proxy is a lightweight solution that dynamically applies CQL2 filters based on user authentication, enabling fine-grained access control without modifying core STAC APIs. This is especially valuable for organizations managing mixed-access datasets, such as public vs. private imagery, role-based access for internal users, or commercial licensing controls. Designed to work seamlessly with pgSTAC, TiTiler, and other STAC-compliant backends, STAC Auth Proxy simplifies the challenge of access control while keeping deployments adaptable. We’re inviting broader adoption and contributions—if you’re working with STAC and need controlled access to your catalogs, check out the repo!
Access and control when and where you need it.
5. Xarray on GPUs
Last month, we collaborated with folks from NCAR and NVIDIA during an OpenHackathon event to streamline the flow of data from Zarr stores to NVIDIA GPUs. This work brings us a step closer towards GPU-native data analysis and ML workflows, and involves coordinated changes to various upstream libraries like zarr-python, kvikIO and xarray. We hope to eventually bring these improvements to cupy-xarray and showcase an end-to-end example utilizing NVIDIA DALI pipelines and nvCOMP GPU decompression for faster reads of large datasets like ERA-5 on Zarr. Keep an eye on what is happening in this GitHub repo.
Improving the ecosystem of libraries enabling faster GPU-native workflows.
Join the Fun
These projects started as “what if?” experiments. Small side quests that turned into contributions to open-source ecosystems. While some are still evolving, we’re excited about the possibilities they provide.
We’re sharing our findings because we believe that the best ideas emerge when we collaborate. If something here sparks your interest, we’d love for you to dive in, whether by contributing to the repos, testing new approaches, or sharing your own insights. Let’s build the future of geospatial tech together. 🚀
We're thrilled to share our ideas with you!
What we're doing.
Latest