Of the 640,000 people released from prison in the United States every year, 75 percent remain unemployed a year later. Formerly-incarcerated job applicants are often denied jobs at the background check stage. Time in prison can also mean gaps in traditional job experience, hurting the applicant’s chances. The Restorative Record strives to enable justice-impacted individuals to showcase their unique stories, rehabilitative efforts, and non-traditional work experiences.
Towards this goal, we are introducing an LLM-based writing assistant to help individuals craft their Restorative Record platform.
While justice-impacted individuals often gain exceptional experience while incarcerated (for example, by spending thousands of hours taking online courses), it can be difficult for employers to assess this experience. The LLM-based writing assistant helps applicants convey the details of their experiences at a critical time.
The one feature of the tool is a simple but effective piece of writing advice: be specific. The LLM tool we designed took as input an applicant’s current writing, and identified sentences that could be made more specific. The tool would then ask follow-up questions to help the applicant elaborate.
For example, consider an applicant who wrote the sentence, “I took many online courses, which inspired me to pursue my passions.” The LLM tool provides the following two questions in response: “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?”
The solution builds on an MIT study that demonstrates how algorithmically-generated grammar and stylistic changes to applicants’ resumes improved employment outcomes. The Restorative Record’s solution pushes this idea further by using an LLM to help applicants improve the substance of their response to adverse notice on a background check.
LLMs are a shiny new tool. We wanted to find a way to leverage them for this population while also avoiding their major weakness, hallucinating. For example, while an LLM could generate a profile directly, this could include false information about an applicant. A lot of the AI-generated writing isn’t very informative. The Restorative Record’s writing assistant helps an applicant dig deeper—to convey details that they might not think of including on their own. In this way it centers the individual and their narrative rather than relying solely on surface-level, generalized content that might lack personal depth or accuracy
The LLM tool was developed with Kenny Peng, who interned with the Restorative Record as part of Cornell’s Siegel PiTech PhD Impact Fellowship.
