Inspiration
The inspiration for Recall AI came from a deeply personal place. Watching loved ones, especially my mother, struggle with memory loss due to dementia made me realize how isolating and disorienting it can be, not just for the person experiencing it, but also for their caregivers. There are apps for reminders and trackers, but very few that preserve the emotional connection. I wanted to build something that didn’t just remind, but remembered with them.
What it does
Recall AI is an interactive memory companion powered by Tavus and Supabase that brings personalized conversations to life. It uses AI-generated avatars to engage patients in meaningful, memory-based conversations—like asking about a specific family photo or recalling a joyful moment. It also includes an Emotional Check-In feature that shifts the tone to offer comfort and assess how the user is feeling through gentle prompts. Conversations are stored, analyzed, and used to deepen future interactions, supporting both memory retention and emotional wellbeing.
How we built it
Tavus for the AI-generated video avatar and real-time conversational interface.
Supabase as our backend to store patient profiles, memory cards (images + stories), and emotional check-ins.
React + Vite for the frontend interface, deployed via Netlify.
We also integrated a system for caregivers to upload family photos and voice notes, which are then embedded in the memory recall sessions. The conversations are interactive and embedded directly in the avatar session using iframes and contextual data.
Challenges we ran into
Integrating Tavus API consistently was challenging due to replica/conversation limits and managing session IDs.
Customizing the avatar session to show photos and quizzes within the conversation stream took experimentation.
Supabase errors on Netlify due to missing environment variables slowed deployment.
Ensuring distinct behaviors for "Memory Recall" vs "Emotional Check-In" required fine-tuning the prompt logic.
Accomplishments that we're proud of
Successfully created a working demo where the avatar references specific family memories with context-aware conversation.
Managed to gamify the interaction with quizzes tied to photos, making it both fun and therapeutic.
Built a caregiver flow that allows remote uploading of photos and voice notes into a shared memory bank.
Delivered an experience that feels emotionally sensitive, practical, and scalable
What we learned
API-based AI avatars can feel incredibly personal when combined with real memory content.
Memory care isn’t just technical—it requires emotional intelligence in how prompts and UI are crafted.
Supabase is powerful but must be configured carefully on deployment platforms like Netlify.
Prompt engineering matters: the difference between a generic and emotionally resonant interaction is in how you feed the AI.
What's next for Recall AI
Add voice emotion detection to adapt the avatar’s response based on tone and mood.
Integrate with calendars and medication reminders to create a full care companion.
Develop caregiver dashboards to monitor emotional trends and memory retention over time.
Launch a beta with real families and memory care centers for usability feedback.
Local language and dialect support to expand accessibility across cultures.
Log in or sign up for Devpost to join the conversation.