Leveraging AI to enhance resumes for justice-impacted individuals with the Cornell University School of Industrial Labor Relations’ Criminal Justice and Employment Initiative (CJEI)
Written by Kenny Peng
Understanding the Issue: Employment Barriers for Justice-Impacted Individuals
Each year, 640,000 people are released from prison in the United States, and a year later, 75 percent find themselves unemployed. Formerly incarcerated individuals often face multiple barriers in securing employment after prison, including background checks, and gaps in traditional professional experience.
This summer, I completed a Siegel PiTech PhD Fellowship at the Cornell University School of Industrial Labor Relations’ Criminal Justice and Employment Initiative (CJEI). I worked on CJEI’s Restorative Record Project, a digital platform that enables justice-impacted individuals to showcase their unique stories, rehabilitative efforts, and non-traditional work experiences. During my fellowship, I designed an LLM-based writing assistant to help individuals craft their profile on the platform.
Building the Solution: An AI-Powered Writing Assistant for Resumes
When I joined CJEI, Jodi Anderson Jr., founder of the Restorative Record and my summer mentor, was enthusiastic about how new technologies, like Large Language Models, could impact their work. I spent the first several weeks of my fellowship discussing ideas and learning about their work by reading existing applications, engaging with key players in the system, such as companies that provide online education to millions of incarcerated individuals, and participating in meetings with the Restorative Records’s software development team.
As we further defined the scope of my project, a theme began to emerge: while justice-impacted individuals often gain valuable experience during incarceration— such as spending thousands of hours taking online courses— employers may struggle to assess the significance of this experience. How could we help formerly incarcerated individuals effectively communicate the value of their experience in job applications?
Jodi and I were inspired by an MIT study showing that algorithmic writing assistance improved job applicants’ grammar and style in their resumes, leading to better employment outcomes. We decided to push this idea further by building a writing assistant that would elevate not only the style, but the substance of a resume.
The basis of our tool is a simple yet effective piece of writing advice: be specific. Our LLM tool analyzes an applicant’s current writing, identifies sentences that could be more specific, and then prompts the applicant with follow-up questions to encourage elaboration. For example, if an applicant wrote, “I took many online courses, which inspired me to pursue my passions,” the tool would respond with two questions: “What specific online courses did you take, and how did they influence your decision to pursue your passions? Can you provide examples of the passions you pursued as a result of these courses?”
Kenny Peng
Ph.D. student, computer science, Cornell Tech
Evaluation for Risks and Next Steps
LLMs are a shiny new tool. Through this project, we aimed to leverage their capabilities, while mitigating their major weakness, hallucination. While we could have had the LLM take a more active role in writing applications, this risked generating false information about applicants. Plus, much of AI-generated writing lacks depth. Our goal is for our writing assistant to help applicants dig deeper and convey details that they might not think of including in their resume on their own.
Currently, the Restorative Record Project’s amazing development team is implementing our LLM tool into their platform. We are also planning pilot studies to evaluate the tool’s efficacy in practice and to gather feedback from both applicants and employers. I am excited to continue working with Jodi and the Restorative Record team as we move forward.