Memo | Devpost

Modulate
Neo4j
Tavily
Tavily AI Research; Medical Report.

Inspiration

By the time a neurologist confirms cognitive decline, the patient has been losing ground for nearly a decade. The signals aren't in a brain scan — they're in how someone talks. The pause before a name. The story told twice. The memory that stops coming up. Nobody was listening. We built Memo to listen.

What it does

Memo calls elderly patients daily, builds a living map of their mind as a knowledge graph, and notifies families when memories go dark — with cited clinical research they can bring to a doctor. No app. No hardware. Just a phone call.

How we built it

Vapi handles calls. Modulate extracts voice biomarkers. OpenAI scores cognitive health. Reka cross-validates. Neo4j maps memories as a reasoning graph. Senso provides continuity between calls. Tavily fetches clinical research on anomalies. Yutori finds local care providers. Render runs everything.

Challenges we ran into

We started scoring calls. It worked but wasn't meaningful. The breakthrough: we weren't scoring Dorothy, we were mapping her mind. Rebuilding Neo4j around memory instances and edge decay transformed a filing system into a reasoning engine.

Accomplishments that we're proud of

A personal memory graph built through daily conversation — detecting change not by comparing to averages, but by watching one person's inner world shift over time. The graph reasons autonomously. It works for anyone who can answer a phone.

What we learned

The most powerful signal isn't what someone says — it's what they stop saying. Families don't trust a number. They trust a cited study connected to their mother's specific observed pattern.

What's next for Memo

Clinical validation with a memory care research partner. Institutional distribution through home care agencies. Longer term, the memory graph becomes a data asset powering Alzheimer's drug trials.