IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

AI-Powered Search Drives Intelligent Texas Child Support Tool

A team at the Office of the Attorney General built a search solution to help child support field case workers with a major pain point — time-consuming research. The result demonstrates the “art of the possible.”

A man in an Oxford shirt holds a magnifying glass enlarging the letters "AI."
Shutterstock
When the Texas Office of the Attorney General (OAG) looked at challenges facing child support field case workers, time-consuming research stood out as a major pain point.

Reducing time spent finding the most current legal and practice documentation turned out to be a project that could be done with existing tools, creating a single knowledge base and a robust AI search tool for the OAG’s Child Support Division.

CIO Tina McLeod and Chief Data Officer Antonia Hernandez were on hand during the Texas Association for Strategic Solutions and Collaboration in Computing (TASSCC) 2024 conference earlier this month, where they took a deep dive into creating an intelligent search engine.

One goal was to have queries that would return an accurate answer or a “no answer” response.

Building on the Microsoft Azure Cosmos DB platform, the team took nonconfidential documents and data, cleaned everything up, indexed them, and placed them into a “ring fence solution” for search. The process included creating a glossary of some 150 acronyms, allowing for saved searches and citations, using Active Directory for authentication and making it all mobile friendly.

“The system is monitored and managed using various Azure logging services. But I want you to know we own all the data — very important. Azure might help us transform that data into meaningful and relevant information, but it’s a ring fence solution that we control, so we can’t have humans that are in the system hallucinate responses,” Hernandez said. “So, if you ask, ‘five plus seven equals 15’ over and over again into a non-ring fence solution, eventually, it’s going to start giving you whatever answer it pleases and thinks you want.

“Not here. We can make sure that the machine also cannot inadvertently change or manipulate responses you'll see in our demonstration, we're going to give you three different questions, the last of which is going to show you that there’s no answer available. That might be frustrating to some users who can’t exactly figure out what they want, but it does safeguard us.”

They debuted the search for about 40 employees and have been told there has been about one hour per week saved by each. That could translate to thousands of hours, McLeod said.

The search uses retrieval augmented generation (RAG) because of “ease of implementation,” and source data remains securely in the OAG’s systems. This project can be replicated in other agencies.

The Child Support Division has more than 2,200 employees. It establishes paternity, obtains court orders for financial and medical support, enforces support orders, collects payments, and supports its members along the way.

“Our field case workers spend an enormous (amount of) time on policy, on case and legal research, and it’s expected that they navigate these complexities of procedural and complex landscapes, interpret legal documentation and ensure compliance with state, local and federal regulations,” Hernandez said. “This responsibility demands a substantial amount of their time, which could otherwise be spent on direct casework and in front of the Texas constituents.”

There are nearly 15 million members in the child support system, 1.5 million active cases at any given time and more than $4.5 billion distributed in payments in a year, McLeod said.

“When you think of the breadth and scope of child support ... when you think of the lifespan ... a case could go from zero to 18 years,” she said. “It is an obligation that is never forgiven, so we have cases in our system that can last 40-plus years."

The pair offered these considerations for augmented search:
  • Know what documentation is needed
  • Look at quality control
  • Define and model the data
  • Know the data policy
  • Use analytics to understand use
  • Decide the frequency of documentation updates

“You can imagine the ROI in a piece like this,” McLeod said. “But I think what’s even more exciting is the art of the possible that comes with people’s imagination. It has been a catalyst for them to really imagine that art of the possible.

“And to what Antonia said, I feel you’re only as good as your data, and it’s amazing how all of a sudden, something that was seemingly as boring as data tagging and classification, data cleansing and accuracy has become just front and center of nearly everything we do in a modernization effort.”

This story first appeared in Industry Insider — Texas, part of e.Republic, Government Technology’s parent company.
Rae D. DeShong is a Dallas-based e.Republic staff writer and has worked at The Dallas Morning News and as a community college administrator.