From Prompt to Public Service: Putting Advanced AI Use Cases into Production for Government Teams

Gavin Rozzi presented this workshop as part of CoPPER’s four-part series on AI in state government research, providing state employees with practical demonstrations of how to move AI use cases from experimental prompts to production-ready government services.

Workshop Series Context

This presentation was the culminating session in CoPPER’s inaugural AI workshop series, “Workshops: AI in State Research,” held throughout July 2025. The series featured four Thursday sessions, each showcasing different applications of AI by state employees in their operational roles. Rozzi’s session focused specifically on advanced implementation and production deployment of AI solutions.

Series Overview

The workshop series covered:

July 10: State of AI in the State - Use Cases, Tools, and Policy (Dr. Walker Gosrich, NJ Office of Innovation)
July 17: Automate Anything (Almost) - AI Integration with Power Automate and Microsoft 365 (Brendan Lee, NJ Department of Children and Families)
July 24: Leveraging AI for State Research - Advanced Techniques in Qualitative Data Analysis (Dr. Jory Catalpa)
July 31: From Prompt to Public Service - Putting Advanced AI Use Cases into Production (Gavin Rozzi, NJ Department of Community Affairs)

Each 1.5-hour workshop combined one hour of learning and demonstration with 30 minutes of Q&A, creating an interactive format for peer learning among state employees.

Role and Expertise

Rozzi presented in his capacity as Director of the Division of Housing & Community Resources Data Center at the New Jersey Department of Community Affairs. In this role, he leads division-wide strategy for data governance, analytics, and modern digital services supporting housing, eviction prevention, lead-safety, and other critical programs.

His practical experience includes:

Overseeing an enterprise data warehouse
Managing a portfolio of data products and interactive websites
Leading cloud-native platforms that convert raw information into actionable insights
Building civic technology platforms serving the public directly
Founding open-data projects in New Jersey’s civic technology community

This combination of government technology leadership and civic innovation background positioned him to demonstrate how advanced AI techniques can be responsibly implemented in real government contexts.

Workshop Overview

The session provided a deep dive into state-ready, advanced AI workflows designed to dramatically shorten the path from concept to production. Rather than theoretical possibilities, the workshop focused on practical demonstrations of techniques that state teams can implement within existing security, compliance, and branding requirements.

Key Topics and Demonstrations

Vibe-Coding Websites

Concept and Application

Using modern generative AI models to draft polished, accessible landing pages in minutes
Rapid prototyping for program teams needing public information sites
Maintaining state security and branding standards throughout the process
Balancing speed of development with quality and compliance

Practical Benefits for Government Teams

Program teams can stand up public information sites quickly
Reduces bottlenecks in web development workflows
Enables rapid response to policy changes or new programs
Maintains professional standards while accelerating delivery
Empowers non-technical staff to contribute to web presence

Live Demonstration The workshop included real-time demonstrations of:

Prompt engineering for government-appropriate content
Generating accessible HTML and CSS code
Incorporating state design systems and branding
Ensuring ADA compliance in generated markup
Iterating quickly based on stakeholder feedback

Security and Compliance Considerations

Working within state security frameworks
Handling sensitive information appropriately
Maintaining separation between public and restricted systems
Documenting AI-assisted development processes
Ensuring human review and oversight

Regulation-Aware Chatbots

Building AI Assistants for Policy Questions

Creating chatbots that can answer staff questions using agency-specific policy documents
Integrating internal datasets and procedures
Building in proper citation so answers remain trustworthy
Reducing burden on subject matter experts for routine inquiries

Technical Architecture

Vector databases for semantic search over policy documents
Retrieval-augmented generation (RAG) approaches
Document preprocessing and chunking strategies
Citation tracking and source attribution
Confidence scoring for answers

Use Cases in Government

Employee onboarding and training support
Policy interpretation assistance
Procedural guidance for complex regulations
Quick reference for frequently asked questions
Consistency in policy application

Trust and Verification

Why citation matters in government AI applications
Human-in-the-loop verification processes
Handling edge cases and uncertain responses
Escalation paths for complex questions
Audit trails for AI-assisted decisions

Demonstration Walkthrough Live demonstration of:

Ingesting policy documents and regulations
Querying the system with realistic staff questions
Examining source citations in responses
Testing accuracy against known policy positions
Identifying when the system appropriately defers to humans

Complementary Tools and Guardrails

Cloud Services for Government AI

Azure OpenAI and government cloud options
AWS Bedrock for state deployments
Data residency and sovereignty considerations
Cost management and budgeting
Service level agreements for government use

Prompt-Writing Tactics

Crafting effective prompts for government contexts
Incorporating policy constraints into prompts
Maintaining appropriate tone and formality
Handling sensitive topics appropriately
Iterative prompt refinement techniques

Credential Management

API key security best practices
Role-based access control
Secrets management in government environments
Audit logging for AI service usage
Separation of development and production credentials

Staying Compliant While Experimenting

Working within procurement constraints
Security review processes for new tools
Privacy impact assessments
Risk management frameworks
Balancing innovation with governance

Testing and Validation

Red teaming AI applications
Bias detection and mitigation
Performance monitoring
User acceptance testing
Continuous improvement processes

Value Proposition for State Agencies

Where AI Adds Value Right Now

Immediate Applications

Automating routine content generation
Accelerating web development cycles
Providing 24/7 policy guidance
Reducing repetitive manual tasks
Improving accessibility of complex information

Resource Optimization

Freeing staff for higher-value work
Reducing backlog in web requests
Scaling support without proportional staff increases
Improving consistency across outputs
Accelerating time-to-market for new initiatives

Quality Improvements

More consistent formatting and style
Better accessibility compliance
Reduced errors in routine tasks
Enhanced user experience
Data-driven continuous improvement

Where Human Touch Still Matters

Critical Human Roles

Final decision-making on policy matters
Oversight and quality assurance
Handling sensitive or complex cases
Building trust with stakeholders
Exercising judgment in ambiguous situations

Ethical Considerations

Ensuring equity and fairness
Protecting privacy and confidentiality
Maintaining transparency
Upholding accountability
Preserving human dignity

Organizational Wisdom

Institutional knowledge and context
Understanding political dynamics
Navigating inter-agency relationships
Building coalitions and buy-in
Long-term strategic thinking

Practical Implementation Guidance

Getting Started

Assessment Phase

Identifying high-impact use cases
Evaluating technical readiness
Understanding compliance requirements
Securing necessary approvals
Building support from leadership

Pilot Projects

Starting small with contained scope
Measuring success metrics
Gathering user feedback
Documenting lessons learned
Scaling what works

Team Building

Identifying champions within agency
Building cross-functional teams
Training staff on AI tools
Developing internal expertise
Creating communities of practice

Overcoming Common Barriers

Procurement Challenges

Working within existing contracts
Leveraging state cooperative agreements
Building business cases for new tools
Navigating approval processes
Managing budget constraints

Security and Privacy Concerns

Conducting thorough risk assessments
Implementing appropriate controls
Engaging security teams early
Documenting safeguards
Maintaining compliance

Cultural Resistance

Addressing fears about automation
Demonstrating value through quick wins
Emphasizing augmentation over replacement
Building trust through transparency
Celebrating successes

Workshop Format and Engagement

Live Demonstrations

The workshop emphasized hands-on demonstrations over theoretical discussion:

Real code and real systems
Actual government use cases
Live problem-solving
Transparent discussion of limitations
Honest assessment of trade-offs

Extended Q&A Session

The 30-minute Q&A provided opportunity for:

Addressing specific agency scenarios
Discussing implementation challenges
Sharing experiences across agencies
Building peer networks
Identifying collaboration opportunities

Plain-Language Guidance

Content was delivered in accessible, non-technical language:

Avoiding unnecessary jargon
Explaining concepts through analogies
Focusing on practical implications
Providing concrete examples
Making AI approachable

Connection to DCA Data Center Mission

Organizational Context

The DCA Data Center exemplifies the principles demonstrated in this workshop:

Converting raw information into actionable insights
Building platforms that serve both policymakers and the public
Maintaining high standards for data governance
Delivering modern digital services
Supporting critical programs with technology

Practical Applications at DCA

The techniques presented have direct applications to:

Housing program information sites
Eviction prevention resources
Lead-safety public education
Policy guidance for local officials
Data visualization platforms

Lessons from Production Systems

Drawing on operational experience with:

Enterprise data warehouses
Public-facing interactive websites
Cloud-native platforms
Data products for diverse stakeholders
Systems serving millions of New Jersey residents

Broader Context: AI in New Jersey Government

Statewide AI Initiatives

The workshop connected to broader efforts across New Jersey government:

NJ Office of Innovation’s AI policy development (Dr. Walker Gosrich)
Department of Children and Families’ automation initiatives (Brendan Lee)
Research and evaluation applications (Dr. Jory Catalpa)
Cross-agency collaboration on AI adoption
Shared learning and best practices

Responsible AI Framework

Emphasis on responsible deployment:

Human oversight and accountability
Transparency in AI-assisted decisions
Equity and fairness considerations
Privacy protection
Continuous monitoring and improvement

Building State Capacity

Contributing to statewide goals of:

Modernizing government services
Improving efficiency and effectiveness
Enhancing accessibility
Supporting evidence-based policy
Attracting and retaining technical talent

Impact and Outcomes

Immediate Value for Participants

State employees gained:

Practical techniques applicable to their work
Understanding of implementation requirements
Realistic assessment of AI capabilities
Network of peers working on similar challenges
Confidence to experiment responsibly

Long-Term Capacity Building

The workshop contributed to:

Growing AI literacy across state government
Building community of practice
Reducing barriers to responsible AI adoption
Sharing successful patterns and anti-patterns
Accelerating learning across agencies

Advancing Government Innovation

Supporting broader goals of:

Digital transformation in public sector
Data-driven decision making
Improved constituent services
More efficient operations
Modern, responsive government

Key Takeaways

For Government AI Implementation

Essential Success Factors

Start with real problems that AI can meaningfully address
Build trust through transparency and proper citation
Maintain human oversight for critical decisions
Work within compliance frameworks from the beginning
Measure and iterate based on actual usage and outcomes

Common Pitfalls to Avoid

Treating AI as solution looking for problems
Skipping security and privacy reviews
Over-promising on capabilities
Neglecting change management
Failing to plan for ongoing maintenance

Sustainable Approach

Balance innovation with governance
Build internal expertise
Document processes and decisions
Share learnings across agencies
Plan for long-term sustainability

Technical Insights

What Works in Government Settings

Focused use cases with clear success metrics
Gradual rollout with continuous feedback
Strong governance and oversight
Integration with existing systems
User-centered design

Implementation Realities

Procurement takes time - plan accordingly
Security reviews are essential, not optional
User adoption requires intentional effort
Maintenance and updates are ongoing work
Cross-functional collaboration is critical

Connection to Civic Technology Background

Open Data and Transparency

The workshop reflected principles from Rozzi’s civic technology work:

Making government more accessible
Empowering citizens and staff with information
Building trust through transparency
Using technology for public good
Supporting democratic governance

Platforms Serving Residents

Experience from projects like OPRAmachine informed:

User-centered design for government services
Building systems that serve the public directly
Balancing openness with necessary protections
Scaling platforms to broad populations
Maintaining services over time

Urban Informatics Research

Academic research background contributed:

Evidence-based approach to technology adoption
Rigorous evaluation of interventions
Understanding of sociotechnical systems
Attention to equity implications
Commitment to reproducible methods

Resources and Continued Learning

Workshop Materials

Participants received:

Sample code and implementation patterns
Prompt templates for government use
Compliance checklists
Architecture diagrams
Additional reading and resources

Community of Practice

The CoPPER series established:

Network of AI practitioners across state agencies
Forum for ongoing questions and collaboration
Mechanism for sharing successes and challenges
Connection to statewide initiatives
Support system for innovation

Follow-Up Opportunities

Pathways for continued engagement:

Office hours for implementation questions
Peer learning sessions
Cross-agency collaboration projects
Connection to statewide AI initiatives
Access to updated guidance as field evolves

Significance for Public Sector AI

Bridging Theory and Practice

The workshop addressed gap between:

AI capabilities discussed in media
Practical realities of government implementation
Compliance and governance requirements
Resource constraints and priorities
User needs and organizational capacity

Realistic Expectations

Providing balanced perspective on:

What AI can do well today
Where significant limitations remain
Required investments and ongoing costs
Realistic timelines for implementation
Appropriate use cases vs. over-application

Ethical Implementation

Emphasizing responsible approach:

Human-centered design
Transparency and explainability
Equity and fairness
Privacy and security
Accountability and oversight

Future Directions

Emerging Applications

Looking ahead to:

More sophisticated policy assistants
Automated report generation
Enhanced data analysis capabilities
Improved accessibility tools
Multilingual services

Continued Innovation

Supporting ongoing work in:

Expanding use cases responsibly
Improving AI governance frameworks
Building technical capacity
Sharing best practices
Learning from implementation

Collaborative Development

Fostering:

Cross-agency collaboration
Partnership with vendors
Engagement with research community
Connection to federal initiatives
Learning from other states

This workshop demonstrated practical pathways for bringing advanced AI capabilities into production government services, emphasizing responsible implementation, proper governance, and the essential role of human judgment in deploying AI for public benefit.