WABAC

Catching dementia early, because forgetting shouldn't mean losing.

Comment

FOR SPONSORS/EVALUATORS:

Scroll down for diagrammatic summary of our architecture, key technologies utilized, and sponsor integration success stories.

???? How it all began

42% of Americans aged over 55 will eventually develop dementia, but many digital interventions are inaccessible or overwhelming for this group. We asked ourselves: What if help was as simple as a phone call?

WABAC (say "way-back") is our answer—a memory-holding voice AI caregiver for those who wish to assist people with dementia to relax into handing over memories and getting cognitive support. By integrating Vapi's conversational voice agents with continuous memory monitoring, we aim to provide users and caregivers with a path to clarity, care, and connection.

???? What it does

WABAC is a voice-based AI memory assistant that speaks directly to care receivers and auto-converts their input into actionable information for caregivers.

???? Key Features:

  • ???? Date-Stamped Memory Logging Customers can call and just speak naturally. The assistant captures, dates, and stores their memories and thoughts in a formal manner.

  • ???? Caregiver Dashboard Visualizes memory trends, reminders, and cognitive health metrics. Offers caregivers an open window into the caregiver's loved one over time.

  • ???? Smart Reminders The caregiver or user can schedule events. The agent reminds users in a conversational fashion on subsequent calls.

  • ???? Cognitive Performance Monitoring Monitors memory consistency changes, vocabulary density, and behavior trends to trigger warnings for potential cognitive impairment.

???? How we built it

Our system is simple yet robust—based on accessibility, persistence, and clarity. The diagram below illustrates how data and interaction flow through our system:

WABAC System Diagram <!-- Replace with hosted version of the image -->

???? System Overview

  1. Vapi Voice AI Agents

  2. Manages natural phone-based dialogue

    • Enables care receivers to recall memories and receive reminders
    • Accepts caregiver inputs through agent workflows

  3. Database (Supabase)

    • Stores reminders, memories, and cognitive scores
    • Indexed by user and timestamp for simple querying and trend analysis

  4. Dashboard (Next.js + Chart.js)

  5. Visualizes memory, reminder, and performance insights history

  6. Accessible to caregivers as well as care receivers

???? Challenges we faced

  • Natural data structuring: Extricating memory data that is structured from unstructured, voice-based input without sounding conversational.
  • Voice UX: Intonating agent responses in a friendly tone while still getting key details out.
  • Data modeling: Defining a schema for tracking time, sentiment, and cognitive consistency over time.

???? Boasts of achievement

  • Created a full end-to-end system: voice → memory → dashboard
  • Was able to monitor real-time user memory input via phone and render it
  • Created a product that emotionally connected with mentors as much as with users

Lessons learned

Nicholas: Learned how to think deeply about user experience beyond UI—how something sounds matters when you're developing for care. Vapi and Letta offered us a chance to walk through empathy vs. functionality.

Logan: First time ever with conversational AI and persistent agent memory for me. Taught a lot about system architecture and how to design for actual users who may not be tech-literate.

✈️ What's next for WABAC

Phase 1: Solidify the Core

  • Improve memory summarization
  • Add daily/weekly call summaries for caregivers

Phase 2: Branch Out

  • Deploy to real senior living communities
  • Add support for multiple caregivers per user

Phase 3: Scale

  • Integrate wearables for health signals
  • Collaborate with healthcare professionals to verify cognitive scoring

This is meant to take you through uncovering the technology in our project quickly and concisely.

✅ Sponsor Services Used:

  • Vapi – Call routing, voice AI agents, workflow builder
  • Letta – Conversational context and agentic memory
  • Supabase – User auth, real-time database, data hosting
  • OpenAI GPT-4o – Summarization and extraction of cognitive signals from memory
  • Chart.js + Next.js – Interactive dashboard of caregiver

For additional technical details, refer to our GitHub README.md for:

  • Agent prompt structure
  • Memory schema
  • Demo link for dashboard

Built With

Share this project:

Updates