NIGMS Azure Open AI
Description
There are a number of situations in which administrative activities can be augmented by generative AI, especially when classification of documents is needed but no training data exist.
Detailed example
Input: Text from various components of NIH grant applications. Output: Open AI chat completions (text) or text embeddings (vectors of numbers).
AI / analytics pattern
Generative AI: AI that generates new or synthetic content (e.g., images, videos, audio, text, code).
Automation level / stage
c) Deployed – The use case is being actively authorized or utilized to support the functions or mission of an agency.
Expected benefit
We are using large language models (LLMs), in particular Open AI's models, for business process improvement. We have used these models for visualization of a grant portfolios as well as numerous classification problems: IC prediction, clinical trial prediction, research area prediction, etc. These models allow us to classify documents through simple prompt engineering rather than the laborious process of creating a custom training set from scratch. These models also allow us to reduce the number of applications that humans need to review from tens of thousands of applications to mere hundreds or fewer for a number of tasks.
Controls / human review
ATO: No; PIA: Not published
Data needed
All data used for this project is internal to NIH, mostly administrative data from the NIH IMPAC II database.