Skip to main content
Projects / Demonstration projects / AIthena: Institutional LLM Assisted Biomedical Question Answering

NAIRR Secure: AIthena: Institutional LLM Assisted Biomedical Question Answering

Using the latest in open Large Language Models (LLMs) to power biomedical search, discovery, and understanding from the literature securely and at scale.

The overarching goal of this project is to enable researchers to search and synthesize all biomedical information their institution subscribes to along with publicly available data using the latest in open Large Language Model (LLM) technologies. To enable this, the Oak Ridge Leadership Computing Facility, home of ORNL’s Frontier Exascale supercomputer, will host the “Ask AIthena” application. “Ask AIthena” utilizes state-of-the-art open LLMs to transform biomedical journal articles, pre-print articles, textbooks, clinical databases, etc. into a format that is easy to search and understand and allows users to ask questions from that large text corpus. This approach ensures that copywritten material is never shared or used to train a model and that users are only enabled to search and synthesize the subset of content that their institution subscribes to. “Ask AIthena” provides a simple user interface for non-technical users to ask questions, find relevant citations to their query, and have in-depth discussions about the relevant text corpus to understand a topic thoroughly.

The pilot program will only process publicly available abstracts from PubMed and all full-text articles from Arxiv, Chemrxiv, Biorxiv, and Medrxiv. However, this limitation is purely to demonstrate that the approach is feasible at scale and that different sources of information and publishers can be subset easily per user such that only relevant and provisioned information is returned to users.

Project leads

Sam Michael (NCATS) & Verónica Melesse Vergara (ORNL)

Advancing NAIRR Infrastructure

While the "AIthena: Institutional LLM Assisted Biomedical Question Answering" project does not require a secure enclave because the abstracts and full text papers are public knowledge, it is a valuable proof-of-concept for NAIRR secure because it sets up all of the key infrastructure and computational pieces so that future publishers, institutions, and other partners could submit their institutional affiliations with publishers and be auto-provisioned to appropriate content. Key to this use case is the assurance that only authorized users would be able to see any non-public content, and even these users would only be able to access content their institution had paid for.

Advancing AI research

AIthena combines the best embedding and "chat" models as well as prompting techniques the field has to offer at a scale and scope beyond what most organizations can afford or have the technical expertise to implement. Additionally, it centralizes this resource so that every institution and organization will not perform duplicative and costly work.

Broader Impacts

AIthena's seamless integration and then subsequent scaling on Frontier along with the fine-grained permissions, and secure enclave create a first of its kind use-case for ORNL. Domain experts from across a variety of technical capabilities will be able to utilize Frontier securely and gain insight into the entire breadth of available biomedical literature. This proof of concept could lead to a new search portal for biomedical researchers rather than relying on traditional search engine type queries and results. Researchers could leverage the full power of AI to not just find relevant literature but understand it, integrate it, and more quickly build upon it.

Advancing domain science

This project is truly something that is the tide which helps to raise all biomedical research "ships". Enabling researchers to more quickly and efficiently discover, understand, integrate, and summarize cutting-edge research so that they can take the next step enables the entire field, regardless of focus, to move forward more quickly. We believe that “Ask AIthena” paired with Frontier will transform the way researchers search and find relevant literature and will simultaneously reduce duplicative work across all domains.