Agentic RAG Chatbot
Description
The creation of accurate, detailed, and factually-grounded draft responses to public inquiries takes a large amount of time and could benefit from assistance from LLMs. Because of how complex the responses to the inquiries can be, LLMs alone did not perform well enough to be useful in production, so we decided to build a more advanced chatbot that uses a small team of agents to refine the inquiries, decide what source data to use for grounding, and generate higher-quality draft responses to the inquiries that staff can use as a starting place for writing their replies.
Detailed example
Draft responses to an inquiry to serve as a base for staff. The inquiry draft will follow all CDC public health research, data, and recommendations based on the best science currently available. Inquiry responses will go through a separate agency review process prior to release.
AI / analytics pattern
Agentic AI: AI systems that perform tasks or make decisions autonomously with minimal human intervention.
Automation level / stage
a) Pre-deployment – The use case is in a development or acquisition status.
Expected benefit
Time savings, especially when programs are overwhelmed by acute increases in the volume of inquiries they receive after publishing a new guideline or regulation is the primary expected benefit.
Controls / human review
ATO: Not reported; PIA: Not published