AI-Powered Public Inquiry Chatbot: USDA
The Arizona State University Artificial Intelligence Cloud Innovation Center, powered by Amazon
Web Services (AWS), collaborated with the United States Department of Agriculture (USDA) to
design a proof of concept (POC) AI-powered chatbot that helps the public find accurate answers
across USDA web properties with fewer clicks and clearer next steps. The POC focuses on
delivering sourced responses from USDA-owned content, surfacing citations alongside
answers, and capturing feedback to continuously improve coverage and quality over time.
Problem
USDA receives a high volume of questions through its websites, and users often struggle to
locate the right information across many pages, PDFs, and related domains. Today, the
experience can require significant searching and navigation, increasing friction and delaying
access to critical programs and services.
USDA’s goal for this POC is a centralized, intelligent interface that can:
- Answer common public inquiries accurately using USDA web content
- Provide transparent source citations with each response
- Guide users toward the best “next action” (program pages, applications, resources) in
minimal steps
Student Spotlight
Cloud Developer | Connect with Shaashvat on LinkedIn
Building Ask USDA meant creating a conversational layer over one of the largest bodies of federal government content, and making it feel simple for the person on the other end. Farmers searching for loan programs, ranchers navigating disaster assistance, everyday people trying to understand food safety. These aren't casual searches, and that urgency shaped every decision we made. My work touched the full stack, from the real-time chat infrastructure and retrieval pipeline to the admin dashboard that gives USDA staff visibility into how the tool is actually being used. The biggest takeaway wasn't technical. It was learning what it means to build AI for a public audience where accuracy isn't optional. The hardest part of a tool like this isn't the AI itself. It's earning the user's trust, one well-cited answer at a time.
Full Stack Developer | Connect with Sreeram on LinkedIn
Building Ask USDA was a highly significant project because we were handling data from critical government websites USDA.gov and Farmers.gov. Our goal was to ensure that users could easily access accurate and reliable information from these sources. We leveraged AWS Bedrock Knowledge Base to efficiently retrieve insights from a large collection of web-scraped data, maintaining confidence scores consistently above 90%. I worked closely on validating every user response to ensure it was safe, accurate, and highly relevant. Additionally, we built an admin dashboard that enables administrators to monitor user inflow, feedback, and overall chatbot performance through various key metrics. This was a project I thoroughly enjoyed working on. The chatbot functions as a hovering assistant on the USDA website - a feature I found particularly exciting, as it was my first time building one!
UX Designer | Connect with Ashik on LinkedIn
Working on the Ask USDA project was a rewarding experience, where I designed an AI-powered chatbot to help users quickly find accurate information across the USDA website. The core challenge was reducing friction for users, especially farmers and the general public, who often struggled to navigate complex content and multiple pages. I focused on creating a simple, trustworthy experience that delivered direct answers backed by official sources, while clearly communicating confidence levels and next steps. I also ensured the chatbot felt native to the USDA site by aligning with its design system and keeping interactions non-intrusive, accessible, and easy to use. This project pushed me to think beyond typical chatbot patterns and instead design for clarity, trust, and real-world usability in a government context.
Approach
The CIC team built a serverless web-content RAG chatbot on AWS with three main flows: Visitor Chat, Admin Dashboard, and Knowledge Ingestion.
- AI/RAG: Amazon Bedrock Knowledge Bases to ground answers in indexed USDA web content and return source citations, with a Bedrock foundation model for response generation and embeddings for retrieval.
- Chat streaming: AWS Lambda (API handler) streams responses to the chat widget via Amazon API Gateway (e.g., SSE), supporting the POC requirement for an interactive conversational interface.
- Content ingestion: A web crawler ingests content from usda.gov (and subdomains) and farmers.gov (HTML + PDFs + embedded text). Content is stored in Amazon S3, processed into retrieval-ready chunks/metadata, and indexed into the Bedrock knowledge base.
- Freshness & optimization: A scheduled ingestion job (e.g., Amazon EventBridge + Lambda) performs delta-based refresh (daily change detection, updating only modified content) and applies metadata-driven crawl frequency optimization for fast-changing sections like news/press.
- Frontend: A floating icon chat widget aligned to USDA branding, designed to return sourced answers with visible citations and a confidence level assessment to safely handle uncertainty.
- Low-confidence handling: When confidence is low, the experience falls back to email capture (e.g., Amazon SES) and captures user feedback on incorrect answers for review.
- Admin & analytics: An admin portal backed by Cognito-authenticated APIs reviews flagged responses, routes admin notifications (Amazon SNS), and supports continuous improvement based on feedback.
- Infrastructure: AWS CDK provisions the serverless stack (Lambda, API Gateway, S3, Bedrock resources, and supporting services) for repeatable POC deployment and documentation.
Industry Impact and Problem Solving
This POC is designed to reduce friction for farmers, ranchers, and the public by making USDA’s authoritative information easier to access through conversational Q&A—without sacrificing transparency. By pairing answers with citations and routing low-confidence interactions into a structured feedback path, the solution supports both usability and trust.
“Partnering with ASU’s AI CIC exceeded our expectations. The student team moved with remarkable speed and professionalism, delivering a well-structured prototype in days that turns dispersed USDA content into a transparent, source-backed conversational experience. Their work gives us an actionable foundation for improving public access to USDA information.”
- Fredy Diaz, USDA Deputy Chief Data & AI Officer
Potential for Wider Application
While this engagement focuses on USDA’s public-facing web properties, the same approach can extend to other government agencies and large public-information organizations with complex, frequently updated content libraries. Future expansion could also consider multilingual support and broader accessibility certification efforts beyond the POC scope.
Supporting Artifacts
| Github Link: | Click Here |
Next Steps
We will move the prototype into a short pilot and planning phase: finish reviewing content on USDA.gov sites and finalize which pages to include so the knowledge base uses only authoritative sources, run a timeboxed pilot to validate citations, confidence thresholds, and the feedback loop, and measure user satisfaction and reduction in steps to get answers. Pilot results will inform a phased production plan covering content governance and ongoing content refresh and ownership. Working with the ASU student team reinforced that grounded, source-linked AI combined with human review is a practical, low-risk way to improve public access to USDA information.
-Fredy Diaz, USDA Deputy Chief Data & AI Officer
About the ASU CIC
The ASU Artificial Intelligence Cloud Innovation Center (AI CIC), powered by AWS is a no-cost design thinking and rapid prototyping shop dedicated to bridging the digital divide and driving innovation in the nonprofit, healthcare, education, and government sectors.
Our expert team harnesses Amazon’s pioneering approach to dive deep into high-priority pain points, meticulously define challenges, and craft strategic solutions. We collaborate with AWS solutions architects and talented student workers to develop tailored prototypes showcasing how advanced technology can tackle a wide range of operational and mission-related challenges.
Discover how we use technology to drive innovation. Visit our website at ASU AI CIC or contact us directly at [email protected].