FAIR

EXECUTIVE SUMMARY

Challenge: The New Jersey Division on Civil Rights (DCR) faced an overwhelming surge in civil rights complaints, growing from 393 cases in 2020 to over 1,000 in 2023, creating substantial backlogs and extending wait times for residents seeking justice for discrimination cases.

Solution: FAIR (Fast AI-Assisted Investigation & Review) is an AI-powered tool that streamlines the processing of civil rights complaints by improving transcription quality, standardizing summarization, and enabling interactive information retrieval through an AI assistant.

Impact: The tool has transformed DCR’s ability to process complaints by reducing average time to write interview summaries, streamlining summary format and quality across investigators, providing timestamped events to avoid rewatching interviews, and ultimately reducing overall case resolution times for New Jersey residents experiencing discrimination.

Partners: New Jersey Division on Civil Rights (DCR) and Northeastern University’s Burnes Center for Social Change.

PROBLEM CONTEXT

Background: Civil rights investigations are critical for ensuring equal treatment and justice for all residents. The DCR is responsible for investigating discrimination complaints across employment, housing, and public accommodations sectors in New Jersey.

Urgency/Need: The dramatic 154% increase in civil rights complaints from 2020 to 2023 created a crisis situation where residents faced prolonged delays in receiving justice. Core issues included poor transcription quality from Microsoft Teams, extremely time-intensive manual interview review process, and inconsistent documentation formats.

Target Audience: DCR investigators, legal specialists, and supervisors managing civil rights cases, as well as New Jersey residents filing discrimination complaints who benefit from faster case resolution.

INNOVATION PROCESS

Approach: Human-Centered Design methodology was employed, focusing on understanding user needs and workflow challenges before developing technical solutions.

Co-Creation: Extensive user research included 11 interviews across 2 divisions and 4 positions within the DCR to ensure the tool addressed specific needs of investigators, legal specialists, and supervisors. The solution was designed around the highest-friction point in the workflow: interview review.

Data Sources: Video recordings/transcripts from civil rights complaint interviews, existing case management system data (NJ BIAS), DCR documentation standards and preferred formats, and user feedback and workflow analysis.

AI SOLUTION OVERVIEW

What was built? 

FAIR is a comprehensive AI-powered platform that transforms civil rights complaint processing through improved transcription, standardized summarization, and interactive information retrieval capabilities.

Key Features:

• Advanced Transcription: Uses OpenAI’s GPT4o Transcribe model to convert interview recordings into accurate, searchable transcripts that improve accuracy by 36% compared to Microsoft Teams transcription
• Standardized Summarization: Powered by GPT-4o to generate clear, legally formatted summaries highlighting key allegations, respondent statements, and supporting evidence according to DCR’s preferred documentation structure with easy revision button for additional context
• Interactive Chat Interface: Enables investigators to ask questions about interview content for efficient access to case details without reviewing entire transcripts or rewatching recordings

OUTCOMES AND IMPACT

Quantitative Results:

36% improvement in transcription accuracy compared to Microsoft Teams baseline

Reduced average time to write interview summaries and case documentation from 2+ hours to >5 minutes

Streamlined processing of 1,000+ annual cases

Qualitative Results:

“With this tool, I am able to schedule more interviews. I used to have to pace myself because it would take a lot of time to summarize the interviews”

“Instead of having to dig through the entire interview…AI is able to help me see the full picture”

“very useful, I often ask for specific questions to be asked in interviews…this is helpful to track back to the interview itself, really very helpful”

LESSONS LEARNED

What Worked: Targeted approach focused on highest-friction workflow point (interview review), user-centered design with 11 stakeholder interviews, strong partnership between government agency and academic institution, security-first design with document deletion after processing, and iterative development based on stakeholder feedback.

What Didn’t: Initial scope was too broad attempting AI-powered document generation across 8 user stories which posed risks to legal process integrity.

Adaptability: This solution can be replicated across other civil rights divisions, government agencies handling complaint investigations, and legal organizations requiring standardized documentation and case review processes. The modular architecture allows for customization to different legal frameworks and documentation requirements.

FUTURE ROADMAP

• Integration with DCR Tech Stack: Host tool on DCR’s Azure cloud account and make accessible through NJ BIAS
• User Feedback Mechanism: Add feedback buttons to summary and home pages
• Expand multi-language support for diverse New Jersey communities
• Develop domain-specific models trained on civil rights legal terminology