Government-Scale AI Assistant | DOH x TAMM

From two assistants to one experience. Sahatna previously had its own AI assistant focused specifically on healthcare and wellness, while TAMM AI served as the broader AI assistant across Abu Dhabi government services. This created two separate experiences with different component libraries, interaction patterns, and content models.

Our challenge was to integrate Sahatna AI into TAMM AI so citizens could access healthcare services conversationally — without needing to understand which platform, department, or backend system was responsible for the service.

Designing at the system level. I collaborated closely with the TAMM AI UX team and Malaffi/Sahatna teams to define how healthcare journeys could become part of TAMM AI's wider conversational ecosystem.

Defining one AI interaction language. We started with collaborative workshops between the Sahatna and TAMM AI design teams. Before touching individual screens, we aligned around:

The objective was to understand where the two experiences differed and establish a shared behavioral model. The project followed four major stages, which allowed us to solve the system first and individual journeys second:

From traditional UI to conversational UI. One of the biggest challenges was translating components originally designed for traditional pages into components that work naturally inside AI conversations. Instead of redesigning everything from scratch, we mapped existing TAMM components across:

For every component, we considered: what information survives, what becomes conversational, what should AI explain, what requires explicit selection, what requires confirmation, and when the experience expands beyond chat. We then established behavioral guidelines around these components.

The component work included documented AI design guidelines and mappings back to the original interface so designers and engineers could implement consistent behavior rather than reinterpret the pattern every time.

Redesigning how users interact with services. Traditional digital government services are structured around pages. AI fundamentally changes that interaction model.

Instead of expecting citizens to read a service page and understand what to do, we designed a reusable framework for turning structured services into guided conversations — identifying what citizens actually need at each point in the conversation, rather than simply placing existing webpage content inside a chatbot.

Give users the right information at the right moment. Government and healthcare services can contain significant amounts of information. Dumping that information into chat would simply turn a complex webpage into a complex conversation.

We therefore established content and interaction rules designed around progressive disclosure. Service descriptions were condensed to roughly 350–400 characters, while essential information such as processing time and fees remained visible early in the experience.

AI conversations still need structured interaction. A major design principle was that conversational AI shouldn't mean making users type everything. Different tasks require different interaction modes. We therefore combined:

For example, appointment booking could use conversational input to understand intent while structured components handle physician selection, specialty, facility, date, and confirmation. The appointment journey demonstrates this approach particularly well: physician search, specialty selection, and date selection reuse the same TAMM components used across other AI-enabled services.

Five journeys, one interaction framework. Once the shared architecture was established, we applied it to key Sahatna healthcare experiences. These journeys became part of the wider TAMM AI experience rather than standalone Sahatna chatbot features.

Trust, clarity and control. Healthcare AI introduces different UX responsibilities from general-purpose conversational experiences. Users need to understand what information is being used, what is AI-generated, what action will happen next, when they need to confirm, and how to move from AI to an actual service.

Before accessing health functionality, citizens encounter consent within the conversation, while structured actions use explicit confirmation before progression. The goal was to make AI feel helpful without making consequential actions feel invisible or automatic.

Turning data into understandable next steps. Another important experience was translating complex health records into conversational summaries. Rather than presenting raw medical information as long-form AI text, the experience organizes important information into structured Key Findings with clear next actions.

This combines an AI-generated medical summary with a structured findings card and an actionable path such as appointment booking — an important portfolio point because it demonstrates designing AI not simply for answer generation, but for decision support and task completion.

From five journeys to a reusable AI framework. We deliberately avoided designing patterns that worked only for Sahatna. The service-to-conversation architecture was first tested outside healthcare and then applied to Sahatna, helping validate that the framework could work across different domains.

This meant future AI-enabled services could inherit established behaviors rather than each product team designing its own conversational patterns.