The easy way to get a custom app for your genAI use case.

Meet the AI Observatory:Ā the fastest way to prototype, test, and scale up a custom app for your genAIĀ use case in a secure environment.

AI Observatory
AI Observatory
AI Observatory

How it works

Number 1

šŸ—ļø We build

The AIOĀ team builds a prototype for your genAI use case, including testing and evaluation (T&E)Ā to make sure it's using the best AIĀ model for the job.
Number 2

šŸ”„ You test

You share and test the prototype with your team, and we iterate with you on improving it.
Number 3

šŸ“ˆ We scale

Once everything looks good, we give you Courses of Action (COAs) you can follow to scale up your prototype into a production deployment, including cost estimates.

AIOĀ in action:Ā Accelerating flight test engineers

Hero star

The AIO is building an AIĀ Flight Test Assistant to support and accelerate flight test engineers across every aspect of their operations. This app is being developed in close collaboration with the GenAIĀ team at Edwards AFB.

Flight test cards
Test hazard analyses
Test card app logo

AĀ test card is a full description of a flight test down to the execution level of detail. It includes specific safety limits, named points of contact for the test, and communications procedures between the pilot, coordinator, and engineer. Every flight test needs a test card.

The AIO has developed an app that lets flight test engineers draft test cards in a few seconds for a few pennies of computing time. This has the potential to significantly accelerate flight testing, and free engineers for other mission-critical work.

Test hazard analysis app logo

The Test Hazard Analysis (THA)Ā is a critical step in flight test planning. Unfortunately, each one takes, on average, 4-8 engineering-hours each to draft.

The AIO built an app that reduces THA drafting time down to a few seconds. With this app, it now costs about $2-3 to generate a typical THA draft. DAF CDAO has estimated that this new app alone has the potential to save millions of dollars a year across the Air Force Test community.

Watch Jordan Conner, who leads GenAI implementation for the Air Force Test Center, demonstrate some of the apps in the AIĀ Flight Test Assistant below (best viewed on fullscreen):

FAQ

FAQ dropdown arrow.
Who is the AI Observatory?
The AIĀ Observatory team consists of DAF airmen from the DAF CDAO, the 96th and 412th Test Wings, and Gladstone AI.
FAQ dropdown arrow.
Why should the AIO build my app?
The AIO is purpose-built to quickly prototype and develop genAI use cases. Gladstone has developed an application scaffold with UI and backend components specially designed for genAI user interactions. This lets us quickly develop exactly what you need for your custom use case.

Your app gets deployed inside SAF/CNā€™s Innovation Landing Zone (ILZ), an infrastructure that's been ATOed for CUI with genAI. The AIO gives you metrics and courses of action (COAs) you can use to test and then scale up your app. Accreditation is streamlined by an experienced team adhering to the latest governance guidelines, policies and requirements.
FAQ dropdown arrow.
How do you keep AIOĀ apps secure?
The infrastructure that hosts the AIO'sĀ apps is part of Microsoft Azure's government cloud. Azure GovCloud uses physically isolated data centers located on U.S. soil to ensures a high level of physical and network security compliance. Azure GovCloud offers several LLMs such as OpenAIā€™s ChatGPT as services, and AIOĀ apps use these.
FAQ dropdown arrow.
Will my app be able to access classified data?
AIOĀ apps will soon be cleared for CUIĀ data access. Higher classification levels may roll out in the medium-term future as well.
FAQ dropdown arrow.
Where will my CUI be stored?
AIOĀ apps store CUI in a secure Microsoft Azure IL5 enclave. AIO apps interface with their own logically isolated Azure databases within this enclave.
FAQ dropdown arrow.
Which AIĀ models will my app use?
We use OpenAIā€™s ChatGPT-4o to enable AIĀ capabilities for most use cases. But other commercial and open source alternatives are available, including Anthropic's Claude-3.5 and Meta AI's Llama-3.1. We select alternative AI models based on their performance requirements on your use case and their availability at the IL5 level.
FAQ dropdown arrow.
Do I need to fund my own use case?
Typically yes, but you can reach out even if you haven't secured funding as there may be other options we can discuss.
FAQ dropdown arrow.
What if I've already built a prototype?
Please get in touch!Ā If you've already taken steps towards building an app, then you've probably already developed a good understanding of the intricacies of your use case, what can go wrong with it, and what needs to go right. That means weā€™ll be that much closer to helping you put together a professionally developed and CUI-enabled app, so we can help you get on the path to scaling and sustainment that much faster.
FAQ dropdown arrow.
How long does it take to build an initial app prototype?
Once weā€™ve met to understand your use case, we aim to deliver a prototype app in 1-2 weeks of engineering time. As an example, the AIO delivered an app to automate drafting Test Hazard Analyses (THAs, a multi-million dollar problem across the Air Force Test community) with the first working prototype built in 24 hours. Similarly, the AIO delivered an app to automate drafting Test Cards (detailed checklists for flight tests) in under a week.
Star icon.

Ready to build on the AIO?