Generative avatar & agentic AI orchestration for an elder-care companion - ON-1248

Project type: Research
Desired discipline(s): Engineering - computer / electrical, Engineering, Computer science, Mathematical Sciences, Mathematics
Company: MindAux Inc.
Project Length: 4 to 6 months
Preferred start date: 11/02/2026
Language requirement: English
Location(s): Toronto, ON, Canada
No. of positions: 1
Desired education level: Master'sPhD
Open to applicants registered at an institution outside of Canada: No

About the company: 

MindAux is a Toronto-based AI company building Mira, a multi-language personalized AI companion designed to support older adults, including people living with dementia, in Long-Term Care and retirement home settings. Mira renders a familiar avatar (voice and face reconstructed from recordings and a photo of a family member) and is powered by PAGE - Personalized Adaptive Guided Engagement, a proprietary multi-agent orchestration architecture coordinating specialized AI agents in real time, patent-pending with a US provisional filed April 2026.

MindAux is led by three co-founders: Dan Rico (CEO) contributes 20+ years of AI/computer-vision/robotics R&D leadership, Brandon Lee ((AI & Engineering) brings 10+ years building and shipping AI/ML systems at Meta/Samsung/Uber and Dibyendu Mukherjee (Research) brings 15+ years of research experience, including work at Duke Hospital and a national research trial.

The company's founding mission is personal, rooted in Dan's family experience with dementia. MindAux is currently in active discovery with Toronto-area LTC and retirement homes and has completed the ChaiTech accelerator program at Prosserman JCC. MindAux is a client of IPON - Intellectual Property Ontario and an active participant in Vector Institute's FastLane program.

Describe the project.: 

This project advances two applied research challenges within Generative AI and Agentic AI that are core to Mira's underlying technology: PAGE - Personalized Adaptive Guided Engagement, a proprietary multi-agent orchestration architecture (patent-pending, US provisional filed April 2026) that coordinates specialized AI agents to deliver safe, personalized companionship at scale.

First, generative avatar and voice synthesis:
Mira renders a familiar, personalized avatar - voice and face reconstructed from recordings and a photo of a family member - and current priorities include improving generative video/motion fidelity, natural facial expression and lip-sync accuracy, and reducing latency for real-time conversational rendering across the languages Mira supports.

Second, agentic orchestration:
Mira's backend coordinates multiple specialized agents (conversational, emotional-state monitoring, activity/content-selection) that must operate safely and predictably together, governed by an Emotional Safety Governor that monitors agent outputs in real time to prevent responses that could distress a cognitively vulnerable population. Research goals include improving inter-agent coordination logic, refining the orchestration engine's decision architecture as the agent ecosystem grows, and strengthening the safety-governance layer to scale reliably across a larger, more diverse resident base.

MindAux's main goal is to deliver Mira, a commercial AI companion product for older adults with residents, families, and frontline care staff as the end beneficiaries: residents through personalized connection and reduced isolation, and families and staff through reduced caregiver burden and real-time visibility into resident wellbeing.

Beyond product development, MindAux is seeking to build long-term research collaborations with academic professors, principal investigators, and physicians by running pilots in LTC and retirement homes to validate research approaches, collect data, summarize findings, and publish results at conferences and in peer-reviewed journals.

Two MSc/PhD-level researchers can contribute significantly to addressing the two challenges above, advancing generative avatar/video modeling and agentic orchestration and safety-governance architecture as core building blocks, directly supporting MindAux's upcoming pilots.

Required expertise/skills: 

Strong background in deep learning and generative modeling (diffusion models, GANs, or video/motion synthesis), with experience in talking-head/avatar generation, lip-sync, or facial animation an asset.
For the orchestration track: experience with multi-agent systems, LLM-based agent frameworks, or safety/guardrail design for AI systems interacting with vulnerable populations
Proficiency in Python and modern ML frameworks (PyTorch preferred).
Familiarity with real-time inference optimization and latency reduction is a plus.
Experience or interest in human-computer interaction, healthcare AI, or applied research with elderly or clinical populations is beneficial given the sensitivity of the end-user population.