Artificial intelligence (AI)-assisted development tools are accelerating NASA’s workflows. With access to modern AI tools, NASA’s science teams can pass time-demanding tasks, such as data processing, visualization, and modeling, to AI agents. This essentially buys back time for discovery and allows teams to focus their efforts where they have the greatest impact: analyzing data, innovating new solutions, and steering future work to support NASA’s missions and research projects. 

NASA’s Science Cloud has recently expanded its shared services catalog to offer two new AI-powered development tools: Microsoft’s GitHub Copilot, an AI coding agent that allows users to write code faster and with less effort within a development environment, and the Science Cloud’s Research Platform, a shared, cloud-based notebook environment where users can write and run live code directly in web browsers, with curated AI tools for science workflows.

GitHub Copilot on the Science Cloud

In partnership with Microsoft, the Science Cloud now offers Microsoft’s GitHub Copilot as a project-funded service. Working in a development environment, Copilot simplifies programming tasks by explaining concepts and completing code, proposing edits, and validating files. The Science Cloud manages the operational and security infrastructure (such as identity and access management, operational controls, and subscription management) so that Copilot is ready to use, on-demand, for NASA-funded projects. 

In a recent showcase event, several projects shared how they are using GitHub Copilot to accelerate and improve research pipelines.

NASA POWER Harnesses AI to Streamline Security and APIs

The NASA Prediction Of Worldwide Energy Resources (POWER) project, which provides global NASA Earth observation data to over 1.8 million users via API and web tools, is just one of many projects benefiting from GitHub Copilot on the Science Cloud.

With GitHub Copilot, the POWER team used AI agents to identify and remove dead code and significantly reduce loading times for POWER’s Data Access Viewer, an interactive web-mapping application. This clean-up improved the user experience and enabled quicker access to NASA solar and meteorological data, all while saving weeks of developer time.

Image

This also enabled the POWER team to optimize their code and reduce the size of their Docker image - a blueprint that includes everything needed to run an application, including the code, system tools, libraries, and settings - by half. This change improved data delivery speed by 62%, a performance boost that translated into significant cloud cost savings: operating costs were reduced by 33% while serving 5% more daily requests and delivering 7% more data. These impressive metrics highlight how a single AI-driven code cleanup can deliver a massive return on investment, simultaneously enhancing application performance and driving down operational costs while serving a growing global audience.

AI for Model Training and Manuscript Preparation

Beyond optimizing web applications and reducing operational costs, GitHub Copilot is also helping researchers drastically accelerate their scientific computations. James Warner, PhD, a computational scientist at NASA Langley Research Center, uses AI-assisted development tools for model training and manuscript preparation.

Warner used GitHub Copilot to profile and optimize a model training workflow that originally took 33 hours on four GPUs. After iterating with Copilot for two hours, he achieved a 3.3x speed-up, reducing model training time from 33 hours to approximately 10 hours.

AI was also instrumental in creating and standardizing visualizations for publication. Rather than consuming hours of tedious work reformatting complex figures for an upcoming manuscript, Warner again turned to AI. By providing figure guidelines to GitHub Copilot, Warner was able to create and refactor dozens of publication-ready figures with consistent formatting in just two hours.

Time Required to Create a Complex Visualization with Python

After using the AI-powered development tools, Warner shared several takeaways.. First, reset expectations often. As AI evolves and technology improves, models that once struggled might now be ready to support your workflow. Second, perform verification checks. They are critical for catching errors quickly. Lastly, learn these tools alongside colleagues rather than going it alone. By bouncing ideas off one another and seeing how others use the tool, you'll gain valuable inspiration and motivation to explore exciting new applications in your own daily workflows.

Building on Warner's encouragement to explore new applications, another recent demonstration showcased how these tools can also revolutionize backend infrastructure management.

Infrastructure Modernization via OpenCode 

Alexey Shiklomanov, PhD, from the Global Modeling and Assimilation Office (GMAO), recently demonstrated another use case for AI-powered development with GitHub Copilot: infrastructure modernization via the OpenCode agentic coding tool in the terminal.

In under 10 minutes, Shiklomanov used Copilot to analyze an existing project structure, create a new publicly accessible S3 bucket, generate and upload a test image, produce a public URL to view the image in a browser, then destroy the infrastructure—something he noted would have taken hours if done manually. Ultimately, this rapid deployment highlights how AI-powered tools can drastically reduce the time spent on routine backend tasks, freeing up teams to focus on their primary research.

AI-Assisted Workflows Accelerate Development 

Access to GitHub Copilot has not only accelerated projects on the Science Cloud, but development of the Science Cloud infrastructure itself. In an analysis of AI-assisted human development, Science Cloud developers found that they were able to deliver 11 new features and ~33,000 lines of production code, tests, and documentation in an estimated 40 to 75 engineer-days of human effort—work that would conservatively require 165 to 245 days with a traditional solo-engineer approach.

The AI-assisted workflow reduced cost and calendar time by 65 to 75% while producing higher documentation and test coverage than is typical for projects of this size and timeline. By using GitHub Copilot for development, the Science Cloud team is able to deliver new features faster and more efficiently, keeping pace with the science that the cloud supports.

The Research Platform

In addition to GitHub Copilot, the Science Cloud is developing another tool to reduce research costs and simplify access to cutting-edge AI capabilities. An open-source, cloud-based JupyterHub environment, The Research Platform lets users write and run live code directly in their web browser.

Previously, teams requiring a JupyterHub environment had to manage their own dedicated deployments, dealing with steep costs and operational complexities. The Research Platform eliminates that burden by providing streamlined access, with identity management, operational controls, and AI governance handled by the Science Cloud team. Because it is a shared, multi-tenant environment, infrastructure costs are distributed so projects only pay for the compute they actually use. Though still in beta, the platform is already proving successful, helping users accelerate their data processing and analysis while drastically reducing overhead costs.

AI-Assisted Tools on the Research Platform

The Research Platform offers a curated set of tools and extensions tailored for scientific research, data analysis, and AI-assisted workflows. Chief among these is Notebook Intelligence (NBI), a built-in large language model (LLM) assistant embedded directly within JupyterLab. NBI provides natural language support for coding, data exploration, and documentation. It features chat, code completion, and an agent mode for interactively creating, editing, and executing notebooks, while also providing a framework to build custom AI tools.

To extend these capabilities, the platform enables on-demand Model Context Protocol (MCP) server deployment. MCP servers are specialized programs that act as secure bridges connecting generative AI models to external data sources, tools, and systems. They serve as a controlled interface, allowing users to discover new AI tools, launch MCP servers, and interact with agents directly from their notebooks.

Beyond NBI and MCP, the platform includes several other capabilities to streamline daily workflows. These include Bucket Explorer 2.0, a redesigned interface for browsing private cloud S3 buckets and public datasets without writing code; NBQueue, a system for rendering resource-intensive notebooks asynchronously on dedicated hardware; Dask integration for managing parallel computing clusters directly within the workspace; GitAccess for visual version control; and detailed Cost and Usage Analytics dashboards that give users full visibility into their compute and storage footprints.

The Future of AI-Powered Tools on the Science Cloud

By accelerating the adoption of AI-powered development, the Science Cloud provides the modern infrastructure necessary to supercharge NASA’s critical work. These tools are designed for anyone who needs a robust computational environment—from writing analysis code and processing massive datasets, to exploring entirely new ways to work.

Whether it is shrinking a 33-hour model training to a single workday, turning hours of manual infrastructure deployment into a 10-minute task, or eliminating the overhead of managing local compute environments, AI is simplifying NASA workflows, ensuring that NASA's most valuable resource—its people—can focus entirely on pushing the boundaries of discovery.

GitHub Copilot and the Research Platform are available to NASA civil servants, contractors, and NASA-funded researchers. To learn more about utilizing these tools for your next project, visit the Science Cloud at science.data.nasa.gov/science-cloud.