Bringing responsible AI into local government

Designing a platform that empowers municipalities to safely integrate AI into their workflows - boosting efficiency while keeping full control over their data.

CONTEXT

The problem

Organizations, especially in the public sector, are increasingly interested in using AI to improve efficiency and automate workflows. However, adoption is often slowed by privacy concerns, integration complexity, customization needs and a limited trust in AI-generated outputs.

This creates a gap between AI’s potential and its practical implementation. Organizations needed a secure, flexible way to integrate AI assistants into their workflows while maintaining full control over data and compliance.

The solution

We designed Mynte, an AI assistant platform that enables organizations to:

Create custom AI assistants (called “metahumans” in phase 1, now we call them assistants).
Provide them with knowledge bases and instructions.
Integrate them into existing systems through APIs.
Maintain compliance with strict privacy and security standards.

The platform supports multiple interaction modes: fully AI-driven conversations, human handoff when needed or hybrid AI and human workflows. This ensures both automation efficiency and reliability.

Key results

Helped shape an AI assistant platform now piloted across several organizational domains, including social services and operational teams
Designed interaction patterns that improved transparency, usability, and trust in AI-generated outputs
Supported stakeholder alignment across multiple teams exploring secure AI adoption

Role

Sole Product Designer

Team

Product Manager, Product Owner, 3 Software Developers, 1 QA Tester

Before

After

Slide between images

Impact & outcomes

While the platform is still evolving, early outcomes show promising adoption and validation. Initial pilot integrations demonstrated the feasibility of secure AI use in enterprise environments. And stakeholder feedback helped us refine priorities and expand the product roadmap. Currently, four other software products within our company have started using the Mynte platform.

DESIGN DECISIONS

My role

I contributed to product discovery and concept validation and my work focused on assistant creation workflows, secure interaction patterns and knowledge base management to ensure reliable and context-aware AI responses.

I worked closely with stakeholders across multiple domains, supporting integration into existing digital products and systems while making sure usability and security are taken into consideration.

This work happened in parallel with other GovTech and AI initiatives within the Innovation team.

Trust & AI transparency

We introduced editable knowledge bases, testing environments and a clear assistant transparency to help build trust and support adoption.

Before

After

Slide between images

[Mynte manager app]
AI assistant profile: before and after iteration based on user feedback.

[Mynte manager app]
AI assistant knowledge base: users can add documents to guide the assistant on what information to use when generating responses.

Customization & scalability

Organizations have different workflows, that is why we designed flexible assistant instructions and the integration option. This helps the platform adapt without redesigning the core experience.

[Mynte manager app]
Assistants can take different roles (chat, transcription or task support) and with customizable tasks, they can adapt to each organization’s workflow.

[Mynte manager app]
Playground view where users can test how the AI assistant they created behaves before deployment.

[Mynte app]
Assistants created by managers can be used across the organization. In this case, the MynScribe assistant supports bilingual communication with real-time transcription and translation.

Workflow automation

We are also exploring workflow automation opportunities. For one client, we are testing how the platform can support an entire workflow starting from an initial client conversation: generating summaries, action plans, documents and product suggestions based on that first interaction (a conversation between an app user and its client). This helps reduce manual work while ensuring people remain in control.

[Custom app]
Testing workflow automation where a single client conversation can trigger summaries, action plans, documents and product suggestions.
The goal is to reduce manual effort while keeping humans in control.

CONCLUSIONS

Flexible platforms enable real-world adoption

Organizations have very different workflows, systems and of course, constraints. Designing configurable assistants and scalable knowledge management helps the platform fit existing environments rather than imposing fixed way of working.
This flexibility is essential for sustainable AI adoption in complex organizational contexts.

Trust drives AI adoption

Organizations are interested in AI, but adoption depends heavily on transparency, security, and the sense of control it provides. Making available clear testing environments, editable knowledge bases and providing transparency around how AI responses are generated helps reduce uncertainty.