Records Redaction Assistant: Tucson PD

The Arizona State University Artificial Intelligence Cloud Innovation Center, powered by Amazon Web Services (AWS), collaborated with the Tucson Police Department (TPD) to prototype an AI-assisted records processing solution for the TPD Records Office. The project focused on helping staff manage public records requests more efficiently by reducing manual intake work and accelerating the redaction of sensitive information in police reports. The prototype was designed to support a guided workflow for records staff, while maintaining human review and privacy protections throughout the process.

The TPD Records Office processes a high volume of public records requests and must carefully redact personally identifiable information (PII) from police reports before release. These workflows were heavily manual, creating delays, increasing data management burden, and raising the risk of accidental disclosure of sensitive information. TPD wanted a solution that could reduce staff workload by streamlining records request handling, improve redaction consistency, and support compliance with privacy requirements.

The CIC team built an AI-assisted records redaction platform on AWS to help the Tucson Police Department streamline public records processing and improve consistency in sensitive information review. The prototype experience is organized around three core flows: Records Request Intake & Process Oversight Case Management, AI-Powered Redaction Review, and Guideline Administration.

• Document intelligence: Amazon Bedrock powers the redaction assistance workflow by analyzing documents, identifying content that may require redaction, and generating structured redaction recommendations aligned to department guidelines. Prompting was designed in multiple stages to improve extraction quality, reasoning, and structured output formatting.

• Content readiness orchestration: AWS Lambda coordinates the end-to-end redaction workflow, including intake validation case creation, document processing, guideline application, redaction suggestion generation, and preparation of reviewer-ready outputs for records staff.

• Records workflow & human review: The platform supports a human-in-the-loop review process where records specialists officers upload source documents, review AI-suggested redactions with supporting rationale, approve or reject recommendations, and generate a final releasable version of the document.

• Guideline administration: Administrators can upload redaction policy or guideline documents, trigger rule extraction, review the resulting structured rules, and manage which guideline set is active for future case processing.

• Data & document storage: Amazon DynamoDB stores case metadata, review decisions, guideline records, and workflow state, while Amazon S3 stores uploaded documents, intake materials, and generated document artifacts.

• API layer: Amazon API Gateway provides secure access between the front-end application and backend services for case handling, guideline management, and document workflow operations.

• Frontend application: A Next.js web application deployed with AWS Amplify provides role-based dashboards and interfaces for records specialists officers and administrators, including uploads, review screens, document review pipeline case tracking, and guideline management.

• Authentication & access control: Amazon Cognito manages secure user authentication and role-based access for specialists officers and administrators interacting with the system.

• Infrastructure & deployment: AWS CDK was used to define and deploy the solution as infrastructure as code, making the prototype easier to provision, manage, and replicate in Tucson PD’s AWS environment.

This solution helps the Tucson Police Department improve how public records requests are reviewed and prepared for release, especially in workflows where accuracy, privacy, and timeliness are essential. By providing an AI-assisted, human-reviewed redaction process, the prototype helps:

• Reduce manual effort involved in reviewing and redacting sensitive police records
• Improve consistency in how redaction guidelines are applied across cases
• Support faster turnaround for public records request processing
• Lower the risk of accidental disclosure of sensitive or personally identifiable information
• Give records staff clearer justification for suggested redactions through guideline-linked reasoning
• Strengthen administrative oversight through structured case tracking and guideline management

The Records Redaction Assistant project represents a long-standing organizational priority to streamline the release of reports to the community, and its advancement marks a critical step forward. By strengthening transparency and reinforcing our service-oriented approach, this initiative delivers clear, measurable value to the greater Tucson community. It enables increased output and improved efficiency while maintaining current staffing levels through the strategic adoption of advanced tools.

- Molly Wise, Police Records Superintendent

While this prototype is focused on the Tucson Police Department’s public records and redaction workflow, its design supports expansion across other law enforcement, public sector, and compliance-heavy document review environments.

The solution’s modular approach enables:

• Adaptation for other police departments and public safety agencies handling records disclosure requests
• Extension to government records offices, legal teams, and compliance units that manage sensitive document review and release
• Reuse of the guideline-management framework for organizations with evolving redaction, disclosure, or privacy rules
• Expansion into adjacent workflows such as FOIA/public records processing, internal investigations, legal discovery, and case file review
• Continued refinement of human-in-the-loop AI review patterns where accuracy, accountability, and policy traceability are critical