Prioritizing patient empathy, safety, and strict regulatory compliance above all else, Amii engineered a novel multi-module architecture that allowed My Viva to transform casual conversations into structured clinical data and scale high-touch care.

Challenge

Chronic disease management requires frequent coaching to sustain patient motivation and scaling high-touch care without losing patient empathy is a critical health industry challenge, as is ensuring responsible data management and patient safety. For My Viva, an Edmonton-based digital health company, scaling this care presented a paradox: human check-ins are too resource-intensive to meet daily demand, while standard automated forms feel mechanical—risking low engagement and compromised health data.

Approach

My Viva partnered with Amii’s machine learning scientists to solve the complex engineering challenges of deploying reliable generative AI in a highly regulated space. Utilizing LangChain and OpenAI’s GPT-4o, the collaboration engineered a multi-module architecture that captures accurate, professional health records through accessible, natural dialogue rather than rigid forms, transforming My Viva’s virtual coach, Yaro®, from a rigid, reactive tool into a proactive, personalized companion capable of delivering continuous behavioral support to thousands of patients simultaneously.

The primary goal of Yaro is to provide patients with non-judgmental accountability and reflection support, while seamlessly capturing accurate professional health records. Through connected data sources and cognitive behavioral therapy principles, patients can reflect on their daily habits and understand how their lifestyle choices directly affect their health outcomes.

• Dr. Nathaniel Maeda

To address these challenges, the project team targeted two primary industry bottlenecks. First, standard AI assistants relying on basic Retrieval-Augmented Generation (RAG) proved too transactional to maintain long-term patient context, track ongoing goals, or build emotional continuity. Second, collecting vital daily reflections like sleep and stress levels traditionally required rigid clinical forms. My Viva needed a solution that could seamlessly extract these critical metrics directly from casual, natural conversations without disrupting the patient experience.

Recognizing that traditional single-model chatbots quickly falter under complex clinical nuances, Amii’s technical team engineered a custom, multi-layered architecture. Rather than relying on a single model to process an entire interaction, Amii restructured the AI's workflow into a pipeline of four specialized, sequential micro-services.

“To hit the performance goals around conversation quality, we settled on using an integrated four-module architecture,” says Dave Staszak, Amii’s lead machine learning scientist. “This approach allowed each module to go deep into a focused task and not get burdened by balancing priorities at once. It also created a flexible architecture where the client could easily add additional features later.”

By embedding behavioural science directly into code, this framework bridges the gap between patient engagement and clinical safety through four core strategies:

Building Trust Through Dialogue: Incorporates clinical coaching techniques (such as OARS) to deliver supportive language that drives sustained user engagement.

Eliminating Liability: Locks parameters to ensure predictable, reproducible responses, removing hazardous AI "creativity."

Enforcing Strict Boundaries: Applies hardcoded algorithmic guardrails that automatically deflect requests for medical diagnoses or treatment plans.

Organic Personalization: Weaves tailored health topics into casual conversations, allowing the AI to guide users proactively while preserving narrative flow.

Impact

My Viva is currently expanding their pilot with an enhanced model built from their Amii engagement. While the company is still testing, the team is happy with the results so far and confident that they have a foundational base to continue to scale their platform, with the full time hire of their ML resident, Moja Ardestani.

“Being able to work with Moja and hiring him full-time after our engagement has been great,” says Dr. Maeda. “Because he’s worked on the project from the beginning, he knows our platform and our business well. He’s been able to expand our team’s technical capabilities and enhance our model so we can continue to scale it responsibly.”