Public Sector N. America

TV News Archive Segmentation

This AI-powered tool automatically transcribes, segments, and labels television news broadcasts, making large media collections more discoverable and accessible to researchers. Developed by the first student cohort of the Vanderbilt Cloud Innovation Lab for Libraries and Applied Digital Preservation, the solution reduces costs by up to 80 percent compared to manual processes while outperforming manually curated data in accuracy.

The Vanderbilt Television News Archive holds approximately 65,000 hours of content, including 1 million news segments and 500,000 commercial breaks recorded continuously since August 5, 1968. Arranging this collection has traditionally been a labor-intensive manual process. The VCIL-developed tool uses AI to transcribe video broadcasts, break up content into time-stamped segments, and label each segment by type, such as news report, commercial, or teaser.

The open-source solution enables other institutions with extensive media collections to adopt the same approach. Future iterations will expand into automated video clipping, metadata generation, and cloud-based collection management tools for archivists.

AWS services used:

  • Amazon Bedrock
  • Amazon Transcribe
  • Amazon S3
  • AWS Lambda

More information:

Automated Knee Alignment Measurement

This AI-powered tool automatically identifies anatomical landmarks on knee X-rays and calculates the measurements surgeons need before performing knee replacement surgery. Built for the University of Pittsburgh School of Medicine, the solution reduces a 15-minute manual task to seconds while achieving a mean error of less than one degree.

Before knee replacement surgery, surgeons must measure precise angles from long-leg radiographs to guide how they align the prosthetic joint. The tool uses a U-Net convolutional neural network trained on over 300 manually annotated radiographs to detect eight key anatomical landmarks, then automatically calculates the Lateral Distal Femoral Angle (LDFA) and Medial Proximal Tibial Angle (MPTA). Rather than simply returning numbers, the solution overlays predicted landmark positions and measurement lines directly onto the X-ray, giving surgeons a quick visual check before accepting the output.

With more than 700,000 knee replacements performed annually in the United States, the potential time savings are substantial. The same approach could be adapted to other procedures requiring manual measurement from medical imaging.

AWS services used:

  • Amazon S3
  • Amazon SageMaker
  • AWS Amplify
  • API Gateway
  • Lambda

More information:

Survey Analysis Agent

This GenAI-powered conversational interface allows users to query survey data in plain English, transforming how organizations extract insights from fan feedback, customer responses, and qualitative data. Developed with Pitt Athletics, the solution eliminates the delays of traditional survey analytics workflows by letting staff simply ask a question and get an answer.

The solution consolidates structured survey responses, open-ended text, sentiment scores, and timestamps into a unified, searchable dataset. Using retrieval-augmented generation (RAG) and embedding-based semantic search, it generates dynamic summaries, explains sentiment trends in context, and surfaces patterns that static dashboards might miss. Each insight includes citations with original response IDs, giving users full visibility into the evidence behind every answer.

Beyond surveys, the underlying approach has broad applications for customer experience teams, donor feedback programs, campus climate surveys, and clinical feedback systems.

AWS services used:

  • Amazon Bedrock
  • Amazon S3
  • AWS Lambda
  • Amazon API Gateway

More information:

Building Permit Review

This solution is an AI-enabled document-intelligence solution capable of reviewing, extracting, and validating information from complex regulatory and engineering documents.

Building permit applications often require extensive documentation and verification. For example, applicants need to prepare multiple types of complex documents containing diagrams, charts, and text, while reviewers must verify hundreds of checklist items against laws and regulations. This collaboration focused on understanding how AI could streamline the review of pool and spa site plans documents that must conform to engineering, zoning, environmental, and safety standards with the goal of reducing manual review time and improving consistency across submissions.

AWS services used: Amazon Textract, Amazon Bedrock, Amazon Comprehend

More information + open-source code

cartoon family at a household pool with desert scenery

Public Inquiry Chat

This AI-powered chatbot helps the public find accurate answers across an organization’s 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.

This solutions is a serverless web-content RAG chatbot built on AWS with three main flows: Visitor Chat, Admin Dashboard, and Knowledge Ingestion

AWS services used: Amazon Bedrock Knowledge Bases, AWS Lambda, Amazon API Gateway, Amazon EventBridge, Amazon SES, Cognito

Realistic cartoon of a young very skinny slim Latino man in his early 30s wearing a t-shirt under farmer overalls, sitting at a computer in a cozy kitchen facing the viewer. Very thin arms, no muscles. Tan skin, dark hair. Computer screen faces the viewer showing green farmland. Above his head a thought bubble reads "!!" Warm kitchen lighting. Wide landscape.

Presentation Analyzer

Human coaching does not scale easily. This solution helps address a common communication challenge in higher education and professional development: presenters often receive limited, inconsistent, or overly general feedback. By creating a structured and repeatable presentation practice experience, this solution supports more personalized coaching and better preparation for high-stakes presentations.

AWS services used for this solution: Amazon Bedrock, Amazon Rekognition, Amazon Transcribe, AWS Lambda, Amazon API Gateway, AWS Kinesis Video Streams, Amazon S3, AWS Cloud Development Kit

cartoon woman speaking to people in audience

Disease Concept Model Builder

In qualitative health research, transforming hours of patient and caregiver interviews into structured Disease Concept Models is a labor-intensive process critical for regulatory submissions and clinical outcome assessments. This solution leverages generative AI to automatically extract and classify key observations — symptoms, patient impacts, and caregiver impacts — from transcribed interviews (VTT format) into meaningful subdomains.

The result is a structured, publication-ready Disease Concept Model that captures the lived experience of a condition from both patient and caregiver perspectives, dramatically reducing manual coding time while supporting peer-reviewed research and patient-centered product development.

AWS services used:

  • Amazon Bedrock
  • Amazon S3
  • AWS Lambda
  • Amazon API Gateway

Disease Concept Model Builder

Diverse Learner Support

The K-12 Co-Teacher is an AI-powered assistant that streamlines the management of student accommodations and support plans for educators. By integrating IEPs, 504 plans, and psychological reports with daily lesson planning, the solution reduces administrative burden while ensuring consistent support for students. The system provides real-time accommodation surfacing, class-level and individual student views, and maintains living records of student progress. Through intelligent analysis of lesson content, it automatically flags potential challenges and recommends inclusive strategies aligned with documented supports, making differentiated instruction more manageable for teachers dealing with multiple class sections.

AWS services utilized:

  • Amazon Bedrock (Anthropic Claude Sonnet 3.7)
  • AWS Amplify
  • Amazon Cognito
  • Amazon API Gateway
  • AWS Lambda
  • Amazon DynamoDB
  • Amazon API Gateway WebSocket APIs

This solution transforms classroom support management by reducing teacher workload, improving consistency of accommodations, and enabling more time for direct student engagement.

Empowering Teachers to Serve Diverse Learners with AI

Sizzle Reel

robot on a pop art background

PDF Accessibility

Many organizations have document collections containing hundreds of thousands of PDF documents, many of which do not meet the Web Content Accessibility Guidelines (WCAG) 2.1 Level AA standards, making it difficult or impossible for individuals relying on assistive technologies to access those documents. To address this issue, the ASU Cloud Innovation Center developed an innovative, artificial intelligence-driven solution designed to remediate documents. Some readily available remediation solutions cost $3-$15 dollars per page, but with this solution, expenses for AWS services are less than a penny per page.

AWS services used:

  • Amazon S3: Used to securely store and manage the documents being remediated
  • AWS Lambda: Automates the file processing workflows
  • ECS (Fargate): Handles document processing efficiently
  • AWS Step Functions: Coordinates the various processes involved in splitting, processing, and merging documents
  • Amazon Bedrock: Generates alt text for images and charts using advanced LLM capabilities

This solution also integrates Adobe Auto-Tag APIs which are designed to automatically clean metadata, apply appropriate tags, and further enable document remediation. 

More information:

An illustration of a robot looking at a piece of paper