McCoy College students create AI tools to support law enforcement and help fight child exploitation

Aeslyn Broughton and Shalim Castro present their audio transcription tool to WilCo guests.

Computer information systems major Aeslyn Broughton, another team leader and WilCo intern, along with her partner, Shalim Castro, created Virgil. This natural language processing tool transcribes and analyzes audio, specifically prison phone calls. "It was designed to listen to hours of in-jail phone calls and flag areas of interest as opposed to having them sit and listen to tens of thousands of hours of audio," she said. "It's not practical, and it chokeholds investigations."

While the district attorney's office currently has an audio transcription tool, it is not AI-enhanced and not nearly as efficient as the students' tool. Broughton said she developed the tool, which includes her own rule-based algorithm, along with open-source large learning models like BERT and BART.

"At the core, Virgil is just a text analysis tool," Broughton said. Virgil is a language-based model that understands contextual meaning. It could pick up slang, code words, or emotional distress, which could help law enforcement and inmates.

Margo Kvatsabaya, also a computer information systems student, with the help of her team members, Jane Ozah, Aziz Hazim, Jaelen Linne, and Muskan Yadav, led the development of an image comparison tool that utilized several methods to determine similarities between potentially illicit images. The tool enables users to upload folders, which are then analyzed to determine the similarity of the images within them. Her team blurred images by default and proposed an authentication gate to unblur them only for authorized users.

"We created a tool that can help the district attorney's office find images that are similar or very distinct," Kvatsabaya said. "We're giving them what the system outputs. We are making the prediction like this image is 75% similar. But otherwise, are they similar or not? It's [the investigators'] decision on their end because we don't know the range for them what is distinct."

Given Texas law, which states that the number of visual depictions of a child in CSAM can influence the severity of the offense and the corresponding punishment, Kelley pointed out that these tools are especially relevant in analyzing evidence.

"Nobody thought about the fact that this meant somebody had to sit there and look at all this [CSAM] and count," Kelley said. "They ended up with investigators that wouldn't stay six or nine months total. They had a lot of people quit under stress."

Kvatsabaya said she trained the AI model on cat images. "Cats are so different and have a lot of patterns. That's why it's a good choice for training the model."

Three students, Broughton, Kvatsabaya, and Kelley's teaching assistant, Sarah Manion, interned with WilCo. They worked with the department to identify needs and brought insights back to the classroom.

"If we could put one or two interns in at Williamson County so that the students would come down there one or two days a week and learn anything they wanted them to learn," he said. "[Then] bring it back to the class as our next iteration of features to add throughout the semester."

Felicia said she was impressed with the students' work after the presentation. "Having an opportunity to work with the students from Texas State has been awesome," she said. "The knowledge that they have, the energy, the care they have taken in creating these tools to help us fight child predators, I can't describe how appreciative I am."

While the AI suite of tools developed by the students is fully functional, each tool requires additional enhancements before WilCo can fully integrate them into the office’s workflow. Broughton, who has been hired with WilCo full time since presenting the tool in class, is working to refine all three tools for implementation.

"They already did have something similar that recorded the calls and gave you a transcript," Broughton explained. However, their tool would only pick up the keyword "sandwich," for example. Virgil can pick up "peanut butter and jelly" and recognize that it's a sandwich-related phrase. It understands context.

This level of nuance opens the door for other use cases. "It can also help inmates by picking up emotional distress on their end and get them help sooner if they're feeling maybe suicidal or depressed," she said.

Tran said that while the disk imaging system is functional, it still faces challenges with large data volumes and the time required to process them. Kvatsabaya said that even with certain inefficiencies in the image comparison tool, removing unnecessary comparison methods could help speed up the analysis process.

Although the tools are not quite ready for deployment, Felicia said she sees potential and practical applications for all three tools.

In the development of these tools, Kelley said security and ethics were a priority. "The students were never put in a position to see or mishandle sensitive data," Kelley said. "Ethical use was drilled into them from day one."

Kelley said the tools were also developed to be legally admissible.

"It works when they do the image analytics on the tail end. If you click the test, it shows you the math, it shows you the charts, and gives an explanation of what it was for and how it did it. That way, if you need an expert witness to review the material, it secures [evidence] locally. It helps with the chain of evidence."

While AI's capabilities are currently limited in the law enforcement field, Felicia said that as its capabilities evolve, she believes its use will increase in the future.

"New and shiny is not normally what we do in prosecution," she said. "I do think it's a trend that we're going to start seeing because I think it's going to be built out of necessity. [At WilCo], we are still operating like we're a small county, and we just can't do that anymore. The volume is too large. The seriousness of the offenses is just too big."

While the tools are in their early stages of development, they open up possibilities for law enforcement applications — and beyond.

Change beyond the classroom

For Kelley and his students, the WilCo project was more than just a class.

"They've got something that they're deploying in the real world that the kids wrote that will help the attorney's office and the investigators process evidence much more quickly and free up their time to chase other bad people," he said.

Exposure to these high-stakes environments is something few undergraduate programs can provide. Kelley said he hopes it becomes the norm.

"The students who were willing to just come in bleary-eyed, tired, with a migraine, they were up all night getting this thing to work because 'I'm going to help kids' are not going to show up looking that way because they help Deloitte balance a spreadsheet," Kelley said.

The students built something that will fundamentally transform how the legal system handles some of its most sensitive and emotionally taxing cases.

"My kids in one class changed the world," Kelley said. "What did you do when you were in college?" ✯