MemoryBridge

Inspiration

Dementia care is not only about memory loss. Patients can experience anxiety, confusion, agitation, depression, and emotional distress, while caregivers and facility staff often struggle to know what will actually help in the moment.

We were inspired by reminiscence therapy, where familiar memories, music, people, sensory cues, and life stories are used to help patients feel grounded, recognized, and connected. The problem is that this is hard to personalize consistently. A caregiver may know the patient deeply, but a facility staff member or clinician may not. MemoryBridge was built to bridge that gap.

What It Does

MemoryBridge is an adaptive reminiscence care planner for people living with dementia.

The app lets caregivers, family contributors, clinicians, and facility staff access a role-based care view. It uses a patient's approved memories, recent mood patterns, preferred music, family connections, sensory cues, and prior session outcomes to prepare a structured reminiscence session.

For each session, MemoryBridge generates:

A selected memory anchor
A therapeutic goal
A short rationale for why that memory was chosen
A guided narration
Safe caregiver prompts
Recommended supports such as music and familiar people
A scene brief for calming visual support
A session flow that caregivers can follow
An outcome form so future sessions can adapt based on what helped

How We Built It

We built MemoryBridge primarily in Jac, using Jac walkers and graph-style care data to model patients, memories, sessions, users, and therapy planning.

The backend includes walkers for:

Loading patient profiles
Retrieving approved memories
Generating adaptive session plans
Recording completed session outcomes
Supporting role-based care views

The adaptive planner reads patient context, memory data, recent mood states, prior outcomes, and available supports before producing a structured session plan. The frontend demo presents the result as a client-facing care workflow that can be used by caregivers, family members, clinicians, or facility staff.

We also added supportive scene media, browser narration, role-based views, and outcome capture so the product feels closer to a real care tool instead of just a static AI text generator.

Use of Jac and Agentic AI

MemoryBridge uses Jac as the primary development language. The project uses Jac walkers to expose care workflows as actions, and the core planning flow behaves like an agentic system.

The planner does more than call a model once. It performs a multi-step care workflow:

Retrieves patient context
Reviews approved memories
Looks at recent mood and previous session outcomes
Selects an appropriate memory anchor
Builds a structured session plan
Produces narration, prompts, supports, and rationale
Saves outcome data so future sessions can adapt

This creates a closed loop where the system can improve session selection based on actual patient response.

Challenges We Faced

One major challenge was balancing hackathon scope with product realism. Dementia care is sensitive, so we wanted the experience to feel useful without pretending to replace clinicians or caregivers.

We also had to decide how much VR to include. The original idea included immersive memory scenes, but we realized the stronger demo was not the VR itself. The more important contribution was the adaptive planning layer: choosing the right memory, explaining why, guiding the session safely, and learning from outcomes.

Another challenge was working around frontend limitations while keeping Jac central to the project. We ended up using a stable web demo connected to the Jac backend so the final presentation would be reliable.

What We Learned

We learned that agentic AI in healthcare is strongest when it supports human judgment instead of replacing it. For this project, the agent does not diagnose or make medical decisions. It helps organize memory-based care, reduce caregiver burden, and make sessions more personal and consistent.

We also learned that the most valuable AI output is not just generated text. It is a plan that connects patient history, emotional state, caregiver workflow, and follow-up data.