Article

Our brain is the most complex organ in our bodies. It controls our behavior, processes input from our senses, and is home to thought, memory, and feeling—everything that constitutes our sense of self. So it’s no surprise that neurosurgery is considered one of the most challenging medical fields. That’s where Atlas Meditech and its Neurosurgical Atlas comes in. 

The Neurosurgical Atlas is the world’s largest nonprofit organization for neurosurgical education and research. Its vast archive of articles, videos, and other content has made it a premier resource for surgeons, residents, and students. However, with the recent advances in generative AI, the team at Atlas Meditech had an idea to make this resource even more useful: AtlasGPT.

Applying Gen AI to Medicine

AtlasGPT uses OpenAI’s Large Language Model (LLM)—the same generative AI model behind ChatGPT—to provide answers to neurosurgical questions in close to real-time. Rather than being trained on the entire internet, like ChatGPT, AtlasGPT is grounded in content from the Neurosurgical Atlas, the Journal of Neurosurgery, and a selection of highly regarded publications from PubMed Central. 

It takes advantage of retrieval-augmented generation (RAG) technology, which allows it to integrate up-to-date information from external sources—the large store of articles it was trained on—into its responses, increasing its accuracy. 

The result: a stunningly accurate Gen AI tool that allows users to ask questions about neurosurgical topics and get highly detailed answers. On top of that, users can pick different personas, so they can get their answers using precise medical terminology or in layman’s terms that the average person can understand. 

A Team Effort

Atlas Meditech provided the vision, the content, and the resources for the project. To provide generative AI expertise, they partnered with Fabriq, who have extensive experience training AI models on private data sets. 

For everything else, Atlas looked to Foster Made. From developing the infrastructure for communicating with Fabriq’s AI model to designing the UI and developing the front end, we used our considerable expertise to take AtlasGPT from a vision to a reality. We also implemented Stych and Paddle to cover authentication and payment respectively, and coordinated communications between the three entities, ensuring the project stayed on track.

Making the vision a reality

There were a number of design considerations that had to be ironed out during the build process. For obvious reasons, AtlasGPT had to be extremely accurate. But on top of that, it needed to be able to show exactly where it was pulling answers from so doctors and students could verify the results or dig deeper for more context. The citations also show nearby figures, tables and video thumbnails, where applicable. 

To facilitate this feature, we designed a UI layout that enables users to see a panel citing specific sources alongside the AI-generated response. This transparency is critical in building trust and helps ensure that users always know exactly where the information is coming from. 

Another issue that Atlas Meditech had was how to accommodate payments from different countries while ensuring that the appropriate taxes were paid. This required implementing a merchant-of-record service, rather than just taking credit card payments. 

Atlas identified Paddle as a vendor that could meet these needs, and we were able to implement it seamlessly into the final product. Now, whether users are paying in New Zealand, Spain, or anywhere in between, the appropriate taxes are collected and paid to the proper authorities. 

A Successful Release

An extensive beta testing period saw over 6000 users sign up to test the platform, and a team of neurosurgeons vetted the answers to ensure the model was returning accurate results. We worked closely with Fabriq and Atlas to incorporate the enormous amount of feedback generated into the final version of AtlasGPT that was released publicly. 

That version of AtlasGPT, which was released in July of 2024, has been a success. The functionality feedback has been overwhelmingly positive, and it met Atlas’ goal of continuing to provide innovative ways to share its comprehensive neurosurgical knowledge base with the world.